# sddiff_boosting_radar_perception_via_spatialdoppler_diffusion__61efe2be.pdf SDDiff: Boosting Radar Perception via Spatial-Doppler Diffusion Shengpeng Wang1 , Xin Luo1 , Yulong Xie1 and Wei Wang2 1Huazhong University of Science and Technology 2Wuhan University {wsp666, l xin, yulong xie}@hust.edu.cn, wangw@whu.edu.cn Point cloud extraction (PCE) and ego velocity estimation (EVE) are key capabilities gaining attention in 3D radar perception. However, existing work typically treats these two tasks independently, which may neglect the interplay between radar s spatial and Doppler domain features, potentially introducing additional bias. In this paper, we observe an underlying correlation between 3D points and ego velocity, which offers reciprocal benefits for PCE and EVE. To fully unlock such inspiring potential, we take the first step to design a Spatial Doppler Diffusion (SDDiff) model for simultaneously dense PCE and accurate EVE. To seamlessly tailor it to radar perception, SDDiff improves the conventional latent diffusion process in three major aspects. First, we introduce a representation that embodies both spatial occupancy and Doppler features. Second, we design a directional diffusion with radar priors to streamline the sampling. Third, we propose Iterative Doppler Refinement to enhance the model s adaptability to density variations and ghosting effects. Extensive evaluations show that SDDiff significantly outperforms stateof-the-art baselines by achieving 59% higher in EVE accuracy, 4 greater in valid generation density while boosting PCE effectiveness and reliability. The code and dataset will be available on https://github.com/Stellar Esti/SDDiff. 1 Introduction Millimeter-wave radar for all-weather perception is increasingly attracting widespread attention in robotics, computer vision, augmented reality, and autonomous driving [Harlow et al., 2024; Zhang et al., 2022; Xu et al., 2021; Adolfsson et al., 2021]. Point cloud extraction (PCE) and ego velocity estimation (EVE) are crucial pillars of radar perception. PCE acts as a low-level sensory process, extracting fundamental object information from reflected radar signals, including position, reflectivity, and Doppler velocity. Conversely, EVE Corresponding author Sensory process Cognitive process Localization Point Cloud Extraction Ego Velocity Estimation ego-velocity Radar Perception mmwave radar Point Clouds Figure 1: Illustration of our work s objective: simultaneously extracting point clouds and estimating ego velocity to enhance radar perception. serves as a high-level cognitive process, leveraging elemental point clouds to infer the radar s ego velocity. As Fig. 1 shows, comprehensive PCE and accurate EVE form a solid perceptual foundation for downstream tasks, including object detection, simultaneous localization and mapping (SLAM), path planning, and autonomous navigation. Existing studies have started to focus on PCE [Prabhakara et al., 2023; Zhang et al., 2024] and EVE [Pang et al., 2024] separately, yielding promising results. These tasks are typically treated independently, with PCE relying solely on signal intensity and EVE anchored in sparse, chaotic points processed by onboard systems. However, addressing these tasks in isolation may overlook the interplay between radar s spatial and Doppler domain features, potentially introducing additional bias. Specifically, on the one hand, non-object regions may display high reflection intensity due to the notorious multi-path effect [Yataka et al., 2024]. Consequently, relying solely on signal intensity may mislead PCE, impeding the balance between detection density and clutter suppression. On the other hand, sparse, noisy points with poor vertical angle resolution severely degrade EVE performance. One of our key observations is that there is a synergy between PCE and EVE, which can reciprocally evoke potential gains from each other. We conduct a feasibility study using the Coloradar Dataset [Kramer et al., 2022], as shown in Fig. 2(a). The orientation and Doppler velocity of 3D point clouds jointly define a surface, with its precise parameters determined by the radar s ego velocity. Given various points, we estimate these parameters by Random Sample Consensus (RANSAC) [Fischler and Bolles, 1981]. As shown in Fig. 2(b), sparse, noisy points after onboard CFAR Proceedings of the Thirty-Fourth International Joint Conference on Artificial Intelligence (IJCAI-25) Doppler Distribution Inliers Outliers Doppler Velocity Doppler Velocity C.P. Points: 108 L.W. Points: 3150 C.P. Inlier: 25.9% L.W. Inlier: 98.6% Indoor settings C.P. Points: 108 L.W. Points: 3150 C.P. Inlier: 19.4% L.W. Inlier: 84.7% Veclocity Error (m/s) Veclocity Error (m/s) 0.016 0.043 0.002 0.031 0.071 0.062 0.009 0.432Outdoor settings Spatial-Doppler domain representation Radar Cube Radar ADC Data Dense Points & Ego Velocity Input Output Li DAR Cube Spatial Occupancy Doppler Velocity Imprecise Precise Spatial Occupancy Doppler Velocity Figure 2: (a) The upper figure presents Doppler velocity distribution derived from sparse points after onboard CFAR post-processing, while the lower figure displays points warped by Li DAR. (b) Given CFAR-processed (C.P.) points or Li DAR-warped (L.W.) points, velocity estimation errors are observed in both indoor and outdoor settings. (c) The brief illustration of SDDiff, where PCE and EVE are jointly refined through the SDDR Purification Process, sculpting coarse representations into fine ones. post-processing lead to an awkward inlier rate (about 25.9% at a threshold of 0.08). Points warped by dense Li DAR point clouds show over 5 improvement in EVE, with the inlier rate reaching 98.6% indoors. This demonstrates that the synergy offers reciprocal benefits: accurate ego velocity helps filter or refine points aligned with Doppler velocity. In turn, robust and dense PCE can further enhance EVE accuracy. To fully unlock such inspiring reciprocal potential, we introduce a Spatial-Doppler Domain Representation (SDDR) for 3D radar perception. As shown in Fig. 2(c), the radar analog-to-digital converter (ADC) signal is encoded into a 3D tensor that embodies both spatial occupancy and Doppler velocity. PCE and EVE are jointly refined through SDDR Purification Process, sculpting coarse, ghost-prone representations into fine, uncontaminated ones. Inspired by remarkable generative capabilities of diffusion models [Ho et al., 2020], we design a Spatial-Doppler Diffusion (SDDiff) model to generate an enhanced SDDR by effectively aligning spatial and Doppler features. Specifically, we propose two key designs that allow SDDiff to seamlessly tailor to 3D radar perception. On the one hand, naive diffusion model aims to transform multi-modality in latent space, requiring a plethora of sampling steps. This high demand is attributed to sampling from a standard Gaussian distribution, which is well-suited for image synthesis. However, it introduces sampling wastage and ambiguous outcomes for the SDDR purification task. To address this, we design a directional diffusion with radar priors, transforming consistent modality in SDDR. This significantly reduces the sampling steps and facilitates finer spatial occupation. On the other hand, we incorporate iteratively refined Doppler velocity profiles into the model as conditions and apply a physical constraint to bridge PCE and EVE based on the synergy. This enables the model to learn adaptive feature representations that are resilient to density variations and ghost effects. We summarize the contributions below: We propose a novel Spatial-Doppler Diffusion model to sculpt coarse representations into fine ones for 3D radar perception. To the best of our knowledge, this is the first attempt to simultaneously attain dense point clouds and accurate ego velocity from Spatial-Doppler Domain Representation. We design a directional Spatial-Doppler diffusion with radar priors to streamline the sampling. This deeply reduces sampling steps and mitigates ambiguous outcomes. We propose Iterative Doppler Refinement, leveraging the Doppler-consistency of refined spatial occupancy to enhance the model s adaptability to density variations and ghosting effects. Experimental results demonstrate that our method outperforms the previous state of the arts and achieves reciprocal benefits across PCE and EVE. Additionally, we will make our self-collected dataset publicly available to the research community. This dataset facilitates further advancements in 3D radar perception. 2 Related Work 2.1 Traditional Model for Radar Detection In early radar detection, low-pass filtered intermediate frequency (IF) signals from millimeter-wave radar were processed via FFT, angle estimation [Schmidt, 1986; Li et al., 2003; Roy and Kailath, 1989], and Constant-False-Alarm Rate (CFAR) [Nitzberg, 1972], resulting in sparse, artifactladen point clouds. Such sparse points are insufficient for downstream tasks such as state estimation, mapping, and navigation. To generate denser radar points, some research [Lai et al., 2024; Qian et al., 2020] leverages radar motion to construct virtual antenna arrays, similar to synthetic aperture radar (SAR) imaging [Ausherman et al., 1984], to enhance angular resolution. However, these methods depend on precise motion estimation or predetermined trajectory, hindering their translation into real-world implementation. Other Proceedings of the Thirty-Fourth International Joint Conference on Artificial Intelligence (IJCAI-25) studies [Cen and Newman, 2019; Cen and Newman, 2018] concentrate on eliminating clutter points caused by multipath effects to enhance the quality of radar point clouds. However, they are limited to high-resolution mechanical scanning radars and unsuitable for commercial off-the-shelf radar. 2.2 Generative Model for Radar Detection Recently, numerous studies have focused on generating denser radar points through cross-modal supervision from high-resolution sensors, such as Li DAR or depth camera. Guan et al. [Guan et al., 2020] leverages a conditional Generative Adversarial Network (c GAN) [Goodfellow et al., 2014; Sun et al., 2021] to achieve object-level imaging and recover the high-frequency shape of the specific object. To materialize scene-level imaging, milli Map [Lu et al., 2020] and Radar HD [Prabhakara et al., 2023] employ a frame-stacking strategy, combining multiple CFAR-filtered Cartesian map patches or planar Range-Azimuth (RA) maps as U-Net [Ronneberger et al., 2015] inputs, with maps projected by Li DAR points for supervision. Similarly, Zhang et al. [Zhang et al., 2024] apply a diffusion model [Ho et al., 2020] to restore radar RA maps under Li DAR supervision, subsequently generating radar points in bird s-eye view (BEV). However, these intensity-only methods overlook the significant contributions of Doppler features to the spatial point cloud s contextual understanding. To enhance 3D point clouds, Luan et al. [Luan et al., 2024] encode CFAR-processed 3D sparse point clouds into BEV images for diffusion model. Unlike the direct approaches, other studies [Cheng et al., 2022; Fan et al., 2024] predict Range-Doppler (RD) maps, and then use conventional angle estimation to derive 3D points. Nevertheless, they still resort to traditional angle estimation. 2.3 Ego Velocity Estimation for Radar Some methods estimate ego velocity by solving a transformation between two consecutive exteroceptive measurements. General registration techniques, such as ICP [Besl and Mc Kay, 1992], NDT [Biber and Straßer, 2003], and their variants [Censi, 2008; Segal et al., 2009], are often applied for this. However, these sensor-agnostic approaches struggle with noisy radar point clouds, which provide poor point-to-point correspondence. While [Cen and Newman, 2019] introduces a keypoint extraction and data association scheme, it is unsuitable for low-resolution commercial radars. More methods leverage Doppler velocities for 2D ego velocity estimation, typically relying on RANSAC [Kellner et al., 2013; Kellner et al., 2014] or end-to-end networks like Radar EVE [Pang et al., 2024]. However, these approaches are built on sparse and chaotic points processed by the radar s onboard system. In open environments with few points, such methods suffer significant performance degradation or even complete failure. To address these challenges, we propose SDDiff, the first method, to the best of our knowledge, that enables both dense point cloud extraction (PCE) and accurate 3D ego velocity estimation (EVE) directly from the raw ADC data of a single-chip radar. 3 Spatial-Doppler Diffusion Model In this section, we present the Directional Spatial-Doppler Diffusion Model with radar priors in Sec. 3.1 and Iterative Doppler Refinement in Sec. 3.2, followed by an overview of SDDiff model as illustrated in Sec. 3.3. 3.1 Directional diffusion with radar priors Spatial-Doppler Domain Representation We focus on radar s information-rich ADC data rather than the sparse points from onboard systems. To simultaneously achieve dense PCE and accurate EVE, we introduce a Spatial-Doppler Domain Representation (SDDR) designed around two key principles: 1) Capturing both spatial occupancy and Doppler velocity with minimal information loss. 2) Eliminating redundant data to reduce computational cost and memory usage when feasible. Specifically, we transform raw radar data into a 4D tensor C RR A E D via fast Fourier transform (FFT), where the dimensions correspond to range, azimuth, elevation, and Doppler. Higher values in the tensor typically indicate a higher likelihood of a point s existence. For a spatial position sk,i,j = (rk, ai, ej), the Doppler velocity vr k,i,j is uniquely determined. Empirically, the maximum value along the Doppler axis is typically at least 6 greater than the second when a point exists at sk,i,j. Using this property, we extract the indices of peak values along the Doppler axis to determine Doppler velocities as shown in Fig. 3. Subsequently, intensity u and Doppler velocity v are concatenated into a polar SDDR C = [u; v] RR A E 2, where intensity represents the spatial occupancy and the index corresponds to the Doppler velocity. This can significantly reduce the data volume and computational overhead. (Intensity, Index) Spatial Occupancy Doppler Velocity Encode into a representation 1.0 0.8 0.6 0.4 0.2 0.0 Doppler Bins Figure 3: Diagram of Spatial-Doppler Domain Representation. Similarly, Li DAR s spatial occupancy can be represented by wrapping the points into a polar format consistent with the radar configuration. Refining PCE and EVE is defined as the SDDR Purification Process, transforming coarse, ghostprone representations into fine, uncontaminated ones. The conventional diffusion process typically transforms the target distribution into a standard Gaussian distribution over long steps, and subsequently iteratively samples a new target starting from the Gaussian noise given conditional embeddings. However, it requires a plethora of sampling steps and produces ambiguous outcomes for the SDDR purification task. In response, we define a directional diffusion process, starting from an initial state with a homogeneous radar representa- Proceedings of the Thirty-Fourth International Joint Conference on Artificial Intelligence (IJCAI-25) tion. The forward and reverse processes are comprehensively presented and analyzed as follows. Forward process Similar to DDPM [Ho et al., 2020], we construct a parameterized Markov chain {x0, x1, , x T } to model the transition between radar and Li DAR distributions as shown in Fig. 4. Given radar s spatial occupancy u0 aligned with Li DAR s x0, the forward diffusion process is defined as gradually adding Gaussian noise ϵ N(0, I) to the target representation along the direction toward the radar prior u0 as follows: q(xt|xt 1, u0) := N(xt; αtxt 1 + (1 αt)u0, λ2 t I) (1) where αt, λt denote pre-defined weight, variance schedule. Theorem 1. Given the directional diffusion process with control signal u0, sampling the state xt at an arbitrary timestep t are expressed in closed form: q(xt|x0, u0) := N(xt; k=1 αkx0 + (1 k=1 αk)u0, β2 t I) where β2 t = Pt k=1 α2 t λ2 k/ α2 k and αn = Qn i=1 αi. The proof of Theorem 1 is included in the appendices of the supplementary material. Eqn. 1 shows that directional diffusion process starts sampling from the radar prior distribution and diffuses toward the Li DAR SSDR. This facilitates fewer sampling steps and mitigates ambiguous outcomes. Reverse process The key of the reverse process is to estimate the posterior distribution pθ (xt 1|xt, u0) with the radar prior u0. Following most of the literature on generative models [Rombach et al., 2022], we aim to choose a tractable distribution N (xt 1; µθ (xt, u0) , Σθ (xt, u0)) by optimizing the variational bound on negative log likelihood of p(x0|u0). t=1 E [DKL [q (xt 1|xt, x0, u0) ||pθ (xt 1|xt, u0)]] where DKL[ || ] denotes the Kullback-Leibler (KL) divergence. Theorem 2. For directional diffusion, with the forward process defined in Eqn. 2, the posterior distribution of the latent variable is tractable when conditioned on x0 and radar s prior u0, and is given in explicit form as follows: p (xt 1|xt, x0, u0) exp (xt 1 µt(x0,t, u0))2 where µt = αtβ2 t 1 β2 t xt + β2 t αtβ2 t 1 λ2 t αt 1 β2 t u0 + λ2 t αt 1x0/β2 t and σ2 t = λ2 tβ2 t 1/β2 t Derivation details on Theorem 2 are provided in the supplementary material appendix due to space limitations. According to Eqn. (2)(4)(5), the mean µt (xt, u0) of the posterior pθ (xt 1|xt, u0) conditioned by the previous latent state xt and the radar prior u0 is further parameterized as follows: µθ (xt, u0) = 1 αt xt + αt 1 αt u0 + λ2 t αtβt ϵθ (6) Therefore, given homogeneous radar prior u0, the objective function in Problem 3 is simplified as: LSpatial = Ext,u0 λ2 t 2α2 t β2 t 1 ||ϵ ϵθ (xt, u0) ||2 2 3.2 Iterative Doppler Refinement Point cloud density variations across indoor and outdoor environments and multi-path-induced ghost points hinder model generalization. Drawing from Fig. 2, we design Iterative Doppler Refinement to mitigate ghost artifacts while progressively enhancing spatial representations for more accurate ego-motion estimation. Specifically, Doppler velocity, i.e. relative radial velocity vr i,j, is determined by the radar s ego-velocity vego and the target s azimuth ai and elevation angle ej when the target is stationary relative to the ground. vr i,j = [cos ai cos ej, sin ai cos ej, sin ej] vego x , vego y , vego z T (8) According to the Cauchy-Schwarz inequality, the peak height of the distribution surface of vr w.r.t. ai and ej corresponds to the magnitude of the radar s velocity, while the peak position indicates the azimuth and elevation angles of the radar s motion as shown in Fig. 4(d). Considering that static targets dominate the radar field of view in scene-level PCE, we utilize Doppler-consistency as a critic to refine spatial occupancy. Based on Eqn. 8 we train a differentiable model fψ (v) for velocity regulation. Given a progressively refined spatial representation xt, a reduction operation along range followed by softmax is applied to obtain a soft mask Mt. Further, we define a Doppler-consistency loss to jointly refine spatial occupancy and ego velocity. LDoppler = E Mt,vt λ2 t α2 t 1 2β2 t β2 t 1 ||vego fψ (Mt v) ||2 2 where donates Hadamard product, the weights are determined by Eqn. 5. 3.3 Overview of Our SDDiff Fig. 4 presents the overview of our SDDiff. During training, the Li DAR representations are softened with a Gaussian kernel to transition to a probabilistic spatial occupancy x0. Subsequently, they and radar priors u0 are fed into the directional spatial-Doppler diffusion module. We adopt SDDNet as the denoising model, designed on a standard 3D U-Net architecture with paird up-down blocks incorporating Res Net and attention mechanism. To optimize computational efficiency, radar Doppler profiles v are embedded w.r.t. chirp channels before injected into each 3D U-Net layer for cross-attention. Subsequently, Iterative Doppler Refinement with the latent variable xt and Doppler profile v as inputs, is initialized to mitigate density variations and ghosting effects. The overall training loss for SDDNet is formulated as: LSDDNet = LSpatial + ωLDoppler (10) Proceedings of the Thirty-Fourth International Joint Conference on Artificial Intelligence (IJCAI-25) Doppler Profile T M 1 t+ M t M 0 M Li DAR PCLs (a) Directional Spatial-Doppler Diffusion Frozen Parameter Learnable Parameter Forward Process Ego-Velocity Decoder Reverse Process Element-wise Multiplication (d) Outputs Fine Coarse 0x tx 1 t+ x Tx Radar ADC Data Doppler Embedding 1 0 ) ( , , t+ μ θ x u Project into AE (c) Iterative Doppler Refinement Down Block3D Down Block3D Down Block3D Down Block3D Filter Decoder ( , , ) ego a e v Figure 4: The overview of our SDDiff. where ω denotes the weight balancing spatial and Doppler losses. Notably, only the parameters of SDDNet are optimized during training. The inference process resembles Stable Diffusion, iteratively sampling the SDDR through Markov chains. The refined SDDR is subsequently employed for ego velocity estimation. 4 Experiments Setup 4.1 Dataset We evaluate the proposed method using both the publicly available Coloradar dataset and a self-collected dataset across different indoor and outdoor scenarios. Colo Radar Dataset. We conduct our method on Colo Radar Dataset [Kramer et al., 2022], which includes raw ADC sample data for single-chip radar, dense Li DAR point clouds, and odometry pose information. We use Li DAR point clouds as the ground truth for PCE and velocity solved through odometry for EVE. The Colo Radar contains both indoor and outdoor scenes with a total of 52 sequences. For a fair comparison with other learning-based baselines, we select the same 36 sequences as the training set and others for testing. Self-Collected Dataset. We collected real-world data from diverse environments using the platform shown in Fig. 5 to assess the model s generalization. The dataset comprises 10,371 frames, with 10% used for fine-tuning and 90% for testing. Reliable odometry for EVE ground truth was obtained using Fast-Livo [Zheng et al., 2022]. 4.2 Implementation Details We implement our SDDiff using Pytorch 1.11.0 with CUDA 12.4. The parameters ω of the weighted spatial and Doppler Single-chip Radar (TI AWR1843) Li DAR (Li VOX HAP) (DCA 1000 EVM) Data Capture Card Experiment Scenes Data Collection Platform Processor (Intel Celeron Figure 5: Our customized hand-held data collection platform. loss are set to 0.01. We apply Gaussian filtering with sigma values of [0.2, 0.5, 1] as the filter encoder and perform offline training for the associated decoder. The Ego-Velocity Decoder is trained according to Eqn. 8, with the Azimuth Elevation (AE) Doppler profile as input. The forward process variances are set to constants increasing linearly from α1 = 0.01 to αT = 0.99 and the noise scale λk is set to 0.1. To illustrate the effectiveness of the directional diffusion strategy, we set the sampling step size T = 20. We train SDDNet for 100 epochs on Colo Radar Dataset with Adam W optimizer and a learning rate 10 4. It takes about 5 days to train our model with a machine using three NVIDIA Ge Force RTX 4090 GPUs and Intel Xeon Gold 6226R CPU. 4.3 Evaluation Metric Baseline. We evaluate the PCE performance of SDDiff compared to traditional method OS-CFAR [1988], as well as generative methods RPDNet [2022], Radar HD [2023] and Proceedings of the Thirty-Fourth International Joint Conference on Artificial Intelligence (IJCAI-25) OS-CFAR Radar HD Radar Diff RPDNet Modified 3D-Diff Lab Indoors Outdoors Figure 6: Qualitative results of different methods and ours on the Colo Radar dataset. Radar Diff [2024]. Since no generative models exist for scene-level 3D PCE, we modify the naive diffusion model with radar spatial occupancy as input for comparison with our approach, denoted as modified 3D-Diff. For EVE performance evaluation, we compare our method with ICP [1992], RANSAC [2013], and Radar EVE [2024]. PCE Metrics. We use Earth Mover s Distance (EMD) and Chamfer Distance (CD) to quantify the similarity between the extracted point clouds P and the ground truth Q. Additionally, we adopt the definition of clutter points as provided in RPDNet [Cheng et al., 2022], as follows: Pclutter = {p P, s.t, d (q, p) > τ1, q Q} (11) where d (q, p) denotes the Euclidean Distance between p and q. Similarly, we define the shot Li DAR points as those generated points that can effectively represent the scene. Qshot = {q Q, p P, s.t, d (p, q) < τ2} (12) We define scene representation level and valid point ratio for point cloud generation quality assessment, and effective generation density for density evaluation. Valid Point Ratio (VPR) is defined as 1 |Pclutter|/|P|, reflecting the reliability of the generated point clouds. Scene Representation Level (SRL) is defined as |Qshot|/|Q|, quantifying the effectiveness of the point cloud generation. Effective Generation Density (EGD) is characterized as |(P Pclutter|)/|Qshot|, referring to the density of the point clouds. EVE Metrics. We evaluate EVE using Mean Absolute Error (MAE) and cumulative velocity error density in both indoor and outdoor scenarios. 5 Results and Analysis 5.1 Point Cloud Extraction Results Qualitatively, we showcase the PCE results of various approaches, as shown in Fig. 6. Our method demonstrates superior density and accuracy in PCE compared to alternatives. Quantitatively, we compare the point clouds generated by different schemes on the Coloradar Dataset against the Li DAR ground truth, using metrics EMD and CD. As presented in Tab. 1, points generated by our method exhibit the closest alignment with the Li DAR ground truth. We also evaluate our method on the self-collected dataset to demonstrate its generalization. As shown in Tab. 2, our approach consistently outperforms others when applied to new scenarios. Reliability and Effectiveness Analysis: Different models show variations in the number of points and clutter at different thresholds. To comprehensively assess the reliability and effectiveness of the generated point clouds, we evaluate the clutter rate, representation level, and density of various PCE methods. As illustrated in Fig. 7, 2D methods exhibit high VPR and low SRL due to the lack of elevation information. our method achieves a balanced trade-off between effectiveness and reliability, outperforming the state-of-theart baseline with a 30% improvement in VPR and a 33% increase in SRL. Moreover, our method ensures the highest valid point cloud density compared to other approaches. This is attributed to the reciprocal benefit mentioned in Fig. 2, which enhances the model s robustness to density variations and ghosting effects. 5.2 Ego Velocity Estimation Results As shown in Tab. 3, our method improves EVE by 30% and 59% over the state-of-the-art in indoor and outdoor scenarios, respectively. This is attributed to our PCE process, which Proceedings of the Thirty-Fourth International Joint Conference on Artificial Intelligence (IJCAI-25) Object Dense Detection Methods Classroom Armyroom Hallways Labroom Aspen room Outdoors Longboard EMD CD EMD CD EMD CD EMD CD EMD CD EMD CD EMD CD OS-CFAR [1988] 1.03 1.09 1.06 1.05 1.15 1.18 1.07 1.29 1.19 1.28 1.54 1.37 4.26 3.57 RPDNet [2022] 0.97 0.85 1.31 1.07 1.40 1.10 1.09 1.01 1.30 1.08 1.69 1.24 2.29 1.67 Radar HD [2023] 0.57 0.53 0.77 0.76 1.63 1.29 1.40 1.11 0.85 0.77 3.20 2.28 4.55 3.95 Radar Diff [2024] 0.72 0.54 0.87 0.68 1.46 1.03 1.13 0.83 1.32 0.89 2.14 1.45 4.16 2.73 Modified 3D-Diff 0.58 0.56 0.77 0.77 0.85 0.84 0.90 0.89 0.88 0.88 1.32 1.29 1.20 1.12 SDDiff (Ours) 0.25 0.24 0.31 0.31 0.56 0.64 0.47 0.53 0.38 0.38 0.75 0.80 0.81 0.88 Table 1: Point Cloud Extraction results on the Coloradar Dataset test split. indicates 2D methods for distinction. Object Dense Detection Methods Indoors Outdoors EMD (50%) CD (50%) MHD (50%) EMD (90%) CD (90%) MHD (90%) EMD (50%) CD (50%) MHD (50%) EMD (90%) CD (90%) MHD (90%) OS-CFAR [1988] 0.78 1.01 0.88 1.23 1.62 1.54 2.32 2.21 2.36 4.72 4.57 5.51 RPDNet [2022] 0.96 0.91 0.83 1.46 1.53 1.34 1.51 1.39 1.30 2.63 2.01 2.49 Radar HD [2023] 0.55 0.51 0.36 1.03 0.89 0.64 1.91 1.53 1.57 4.43 3.24 5.32 Radar Diff [2024] 0.54 0.77 0.31 1.22 2.01 1.56 1.63 1.68 1.68 3.35 3.25 3.33 Modified 3D-Diff 0.42 0.41 0.33 1.14 0.98 0.81 0.79 0.76 0.64 2.05 1.16 1.86 SDDiff (Ours) 0.33 0.33 0.21 0.79 0.79 0.44 0.62 0.61 0.50 0.85 0.84 0.75 Table 2: Point Cloud Similarity Comparison at different percentiles on datasets collected from real-world scenarios. 0.0 0.2 0.4 0.6 0.8 1.0 Scene Representation Level (SRL) Valid Point Ratio (VPR) Higher Reliability Greater Effectiveness OS-CFAR RPDNet Radar HD Radar Diff Modified 3D-Diff SDDiff Figure 7: The evaluation results for effectiveness, reliability, and density of PCE. A larger scatter radius corresponds to a higher EGD. provides more accurate and reliable point clouds for EVE. In contrast, other methods degrade outdoors due to relying on the sparse points processed by the data capture card. EVE Methods MAE ( m/s) Cumulative Density of Error/% (Indoors/Outdoors) 0.05m/s 0.1m/s 0.15m/s 0.2m/s 0.25m/s ICP 0.60/0.77 5/6 12/10 17/15 23/20 29/25 RANSAC 0.31/0.57 35/13 55/27 63/37 66/44 67/47 Radar EVE 0.13/0.27 29/8 51/14 66/22 79/36 87/50 SDDiff 0.09/0.11 31/29 59/55 78/74 89/86 95/93 Table 3: The EVE performance results 5.3 Ablation Study We conduct an ablation study to validate the gains introduced by PCE module i.e. Directional Spatial-Doppler Diffusion and EVE module i.e. Iterative Doppler Refinement. As shown in Tab. 4, the PCE without Doppler information incurs a 15% reduction in PCE performance compared to the whole SDDiff. Nevertheless, it still outperforms Modified 3D-Diff, which suffers from ambiguous outcomes due to its diffusion scheme. Moreover, it speeds up inference by 3.13 compared to Modified 3D-Diff, owing to the more efficient sampling prior distribution. The EVE without reliable spatial occupancy leads to a 1.2 and 3.4 increase in indoor and outdoor errors, respectively. This demonstrates that SDDiff effectively harnesses the potential gains from the synergy between PCE and EVE. PCE Module EVE Module VPR (%) SRL (%) EGD ( ) V.E. (m/s) Speed ( ) " % 65.5 69.2 0.98 - 3.13 % " - - - 0.11/0.37 - " " 77.1 79.6 1.17 0.09/0.11 2.57 Table 4: Ablation quantitative results on EVE and PCE. 6 Conclusion This paper introduces SDDiff, a directional Spatial-Doppler diffusion model to simultaneously enable dense point cloud extraction and accurate ego velocity estimation. To reduce sampling wastage and mitigate ambiguous outcomes, we design a directional diffusion with radar priors to streamline the sampling. Additionally, we design Iterative Doppler Refinement to defend against density variations and ghosting effects. The experiments demonstrate that SDDiff enhances EVE accuracy while improving both the effectiveness and reliability of PCE. Proceedings of the Thirty-Fourth International Joint Conference on Artificial Intelligence (IJCAI-25) Acknowledgments This work was supported in part by National Natural Science Foundation of China with Grant 62471194, and the Fundamental Research Funds for the Central Universities. References [Adolfsson et al., 2021] Daniel Adolfsson, Martin Magnusson, Anas Alhashimi, Achim J Lilienthal, and Henrik Andreasson. Cfear radarodometry-conservative filtering for efficient and accurate radar odometry. In 2021 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pages 5462 5469. IEEE, 2021. [Ausherman et al., 1984] Dale A Ausherman, Adam Kozma, Jack L Walker, Harrison M Jones, and Enrico C Poggio. Developments in radar imaging. IEEE Transactions on Aerospace and Electronic Systems, (4):363 400, 1984. [Besl and Mc Kay, 1992] Paul J Besl and Neil D Mc Kay. Method for registration of 3-d shapes. In Sensor fusion IV: control paradigms and data structures, volume 1611, pages 586 606. Spie, 1992. [Biber and Straßer, 2003] Peter Biber and Wolfgang Straßer. The normal distributions transform: A new approach to laser scan matching. In Proceedings 2003 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS 2003)(Cat. No. 03CH37453), volume 3, pages 2743 2748. IEEE, 2003. [Blake, 1988] Stephen Blake. Os-cfar theory for multiple targets and nonuniform clutter. IEEE transactions on aerospace and electronic systems, 24(6):785 790, 1988. [Cen and Newman, 2018] Sarah H Cen and Paul Newman. Precise ego-motion estimation with millimeter-wave radar under diverse and challenging conditions. In 2018 IEEE International Conference on Robotics and Automation (ICRA), pages 6045 6052. IEEE, 2018. [Cen and Newman, 2019] Sarah H Cen and Paul Newman. Radar-only ego-motion estimation in difficult settings via graph matching. In 2019 International Conference on Robotics and Automation (ICRA), pages 298 304. IEEE, 2019. [Censi, 2008] Andrea Censi. An icp variant using a pointto-line metric. In 2008 IEEE international conference on robotics and automation, pages 19 25. Ieee, 2008. [Cheng et al., 2022] Yuwei Cheng, Jingran Su, Mengxin Jiang, and Yimin Liu. A novel radar point cloud generation method for robot environment perception. IEEE Transactions on Robotics, 38(6):3754 3773, 2022. [Fan et al., 2024] Cong Fan, Shengkai Zhang, Kezhong Liu, Shuai Wang, Zheng Yang, and Wei Wang. Enhancing mmwave radar point cloud via visual-inertial supervision. ar Xiv preprint ar Xiv:2404.17229, 2024. [Fischler and Bolles, 1981] Martin A Fischler and Robert C Bolles. Random sample consensus: a paradigm for model fitting with applications to image analysis and automated cartography. Communications of the ACM, 24(6):381 395, 1981. [Goodfellow et al., 2014] Ian Goodfellow, Jean Pouget Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, Aaron Courville, and Yoshua Bengio. Generative adversarial nets. Advances in neural information processing systems, 27, 2014. [Guan et al., 2020] Junfeng Guan, Sohrab Madani, Suraj Jog, Saurabh Gupta, and Haitham Hassanieh. Through fog high-resolution imaging using millimeter wave radar. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 11464 11473, 2020. [Harlow et al., 2024] Kyle Harlow, Hyesu Jang, Timothy D Barfoot, Ayoung Kim, and Christoffer Heckman. A new wave in robotics: Survey on recent mmwave radar applications in robotics. IEEE Transactions on Robotics, 2024. [Ho et al., 2020] Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising diffusion probabilistic models. Advances in neural information processing systems, 33:6840 6851, 2020. [Kellner et al., 2013] Dominik Kellner, Michael Barjenbruch, Jens Klappstein, J urgen Dickmann, and Klaus Dietmayer. Instantaneous ego-motion estimation using doppler radar. In 16th International IEEE Conference on Intelligent Transportation Systems (ITSC 2013), pages 869 874. IEEE, 2013. [Kellner et al., 2014] Dominik Kellner, Michael Barjenbruch, Jens Klappstein, J urgen Dickmann, and Klaus Dietmayer. Instantaneous ego-motion estimation using multiple doppler radars. In 2014 IEEE International Conference on Robotics and Automation (ICRA), pages 1592 1597. IEEE, 2014. [Kramer et al., 2022] Andrew Kramer, Kyle Harlow, Christopher Williams, and Christoffer Heckman. Coloradar: The direct 3d millimeter wave radar dataset. The International Journal of Robotics Research, 41(4):351 360, 2022. [Lai et al., 2024] Haowen Lai, Gaoxiang Luo, Yifei Liu, and Mingmin Zhao. Enabling visual recognition at radio frequency. In Proceedings of the 30th Annual International Conference on Mobile Computing and Networking, pages 388 403, 2024. [Li et al., 2003] Jian Li, Petre Stoica, and Zhisong Wang. On robust capon beamforming and diagonal loading. IEEE transactions on signal processing, 51(7):1702 1715, 2003. [Lu et al., 2020] Chris Xiaoxuan Lu, Stefano Rosa, Peijun Zhao, Bing Wang, Changhao Chen, John A Stankovic, Niki Trigoni, and Andrew Markham. See through smoke: robust indoor mapping with low-cost mmwave radar. In Proceedings of the 18th International Conference on Mobile Systems, Applications, and Services, pages 14 27, 2020. [Luan et al., 2024] Kai Luan, Chenghao Shi, Neng Wang, Yuwei Cheng, Huimin Lu, and Xieyuanli Chen. Diffusionbased point cloud super-resolution for mmwave radar data. Proceedings of the Thirty-Fourth International Joint Conference on Artificial Intelligence (IJCAI-25) In 2024 IEEE International Conference on Robotics and Automation (ICRA), pages 11171 11177, 2024. [Nitzberg, 1972] Ramon Nitzberg. Constant-false-alarmrate signal processors for several types of interference. IEEE Transactions on Aerospace and Electronic Systems, (1):27 34, 1972. [Pang et al., 2024] Changsong Pang, Xieyuanli Chen, Yimin Liu, Huimin Lu, and Yuwei Cheng. Radarmoseve: A spatial-temporal transformer network for radar-only moving object segmentation and ego-velocity estimation. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 38, pages 4424 4432, 2024. [Prabhakara et al., 2023] Akarsh Prabhakara, Tao Jin, Arnav Das, Gantavya Bhatt, Lilly Kumari, Elahe Soltanaghai, Jeff Bilmes, Swarun Kumar, and Anthony Rowe. High resolution point clouds from mmwave radar. In 2023 IEEE International Conference on Robotics and Automation (ICRA), pages 4135 4142. IEEE, 2023. [Qian et al., 2020] Kun Qian, Zhaoyuan He, and Xinyu Zhang. 3d point cloud generation with millimeter-wave radar. Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies, 4(4):1 23, 2020. [Rombach et al., 2022] Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Bj orn Ommer. High-resolution image synthesis with latent diffusion models. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 10684 10695, 2022. [Ronneberger et al., 2015] Olaf Ronneberger, Philipp Fischer, and Thomas Brox. U-net: Convolutional networks for biomedical image segmentation. In Medical image computing and computer-assisted intervention MICCAI 2015: 18th international conference, Munich, Germany, October 5-9, 2015, proceedings, part III 18, pages 234 241. Springer, 2015. [Roy and Kailath, 1989] Richard Roy and Thomas Kailath. Esprit-estimation of signal parameters via rotational invariance techniques. IEEE Transactions on acoustics, speech, and signal processing, 37(7):984 995, 1989. [Schmidt, 1986] Ralph Schmidt. Multiple emitter location and signal parameter estimation. IEEE transactions on antennas and propagation, 34(3):276 280, 1986. [Segal et al., 2009] Aleksandr Segal, Dirk Haehnel, and Sebastian Thrun. Generalized-icp. In Robotics: science and systems, volume 2, page 435. Seattle, WA, 2009. [Sun et al., 2021] Yue Sun, Zhuoming Huang, Honggang Zhang, Zhi Cao, and Deqiang Xu. 3drimr: 3d reconstruction and imaging via mmwave radar based on deep learning. In 2021 IEEE International Performance, Computing, and Communications Conference (IPCCC), pages 1 8. IEEE, 2021. [Xu et al., 2021] Jingao Xu, Guoxuan Chi, Zheng Yang, Danyang Li, Qian Zhang, Qiang Ma, and Xin Miao. Followupar: Enabling follow-up effects in mobile ar applications. In Proceedings of the 19th Annual International Conference on Mobile Systems, Applications, and Services, pages 1 13, 2021. [Yataka et al., 2024] Ryoma Yataka, Pu Wang, Petros Boufounos, and Ryuhei Takahashi. Sira: Scalable interframe relation and association for radar perception. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 15024 15034, 2024. [Zhang et al., 2022] Mingjin Zhang, Chengyu He, Jing Zhang, Yuxiang Yang, Xiaoqi Peng, and Jie Guo. Sarto-optical image translation via neural partial differential equations. In IJCAI, pages 1644 1650, 2022. [Zhang et al., 2024] Ruibin Zhang, Donglai Xue, Yuhan Wang, Ruixu Geng, and Fei Gao. Towards dense and accurate radar perception via efficient cross-modal diffusion model. IEEE Robotics and Automation Letters, 9(9):7429 7436, 2024. [Zheng et al., 2022] Chunran Zheng, Qingyan Zhu, Wei Xu, Xiyuan Liu, Qizhi Guo, and Fu Zhang. Fast-livo: Fast and tightly-coupled sparse-direct lidar-inertial-visual odometry. In 2022 IEEE/RSJ international conference on intelligent robots and systems (IROS), pages 4003 4009. IEEE, 2022. Proceedings of the Thirty-Fourth International Joint Conference on Artificial Intelligence (IJCAI-25)