Driving · 2026-11-02
UFMF: Radar-Guided Kinematic Priors for Real-Time 4D Occupancy Forecasting
Zeping Zhang, Zhuoya Zhao, Robert Lagani\`ere
AI-assisted summary based on the paper. Consult the linked source for methods and limitations. Benchmark numbers appear after editor verification.
TL;DR
UFMF is a real-time 4D occupancy forecaster that keeps an attention-free diagonal state-space temporal core and adds radar-guided velocity-aware dynamic priors (VADP) plus two training-only geometric constraints (world-frame dynamic supervision and class-conditioned vertical-profile supervision). It achieves 122 FPS on a single RTX 4090 and reaches 0.45 fast-displacement mIoU on nuScenes, outperforming recent occupancy world models at lower computational cost.
Why it matters
The authors target dynamic blank-collapse in lightweight BEV occupancy forecasting, where fast actors blur or disappear at longer horizons due to strict latency budgets. By decomposing motion with a radar-guided dynamic prior and enforcing temporal coherence in a world-aligned frame plus class-conditioned vertical profiles, UFMF aims to preserve fast-moving actor support over multiple seconds while meeting real-time inference constraints for downstream planning.
Method
- Use a diagonal state-space temporal core (attention-free) and a channel-to-height lift; keep inference fast while adding a targeted radar-guided branch.
- Velocity-aware dynamic prior (VADP): fuse semantic history deltas with radar Doppler cues to predict horizon-wise displacement fields as linear extrapolation plus a learned horizon-conditioned residual, then transport a compact dynamic latent subspace during rollout.
- Training-only geometric constraints: world-frame dynamic supervision warps predictions to a shared world-aligned frame for edge and temporal-difference consistency, plus class-conditioned vertical-profile supervision to regularize BEV-to-3D lifting.
Limitation
The study is confined to nuScenes; validating the same kinematic priors across additional sensor suites remains an open direction.
Abstract (from arXiv)
Real-time 4D occupancy forecasting requires reliably extrapolating both ego motion and surrounding actor motion under strict latency budgets. Recent occupancy world models improve long-horizon fidelity through generative decoders and 3D volumetric modeling, yet their computational cost makes real-time deployment challenging. Light-weight BEV forecasters meet latency requirements but often blur or suppress fast-moving actors at longer horizons. We present UFMF (Ultra-Fast Multi-modal Forecaster), a real-time forecaster that keeps an attention-free diagonal state-space temporal core while injecting targeted kinematic and geometric structure where lightweight models typically fail. A radar-guided velocity-aware dynamic prior (VADP) decomposes future displacement into a linear extrapolation plus a learned horizon-dependent residual. The resulting per-cell flow fields transport a compact dynamic latent subspace during rollout. Two auxiliary training constraints complement this branch: world-frame dynamic supervision enforces inter-horizon trajectory coherence, and class-conditioned vertical-profile supervision regularizes BEV-to-3D lifting. UFMF operates at 122 FPS on a single RTX 4090 and reaches 0.45 fast-displacement mIoU on nuScenes, outperforming recent occupancy world models at a fraction of their computational cost.
Related papers
- FreeOcc: Decoupling Ego-Motion for Efficient 4D Occupancy Forecasting via Continuous Flow Matching
- InfraWorld: Motion-Aware World Modeling for Roadside 4D Occupancy Forecasting
- Learning the 4D World through Point Trajectory Forecasting
- LO-World: Unified BEV Latent Modeling for Joint 4D Occupancy and LiDAR Forecasting
- WorldFlow: Joint Video and 3D Motion Prediction via Large-Scale Pretraining
Publication record
Bibliographic evidence describes where and when a record appeared. It does not validate scientific claims.
First announced by arXiv: 2026-11-02 · Indexed by Pub-AI: 2026-11-02 UTC
UFMF: Radar-Guided Kinematic Priors for Real-Time 4D Occupancy Forecasting
First announced by arXiv: 2026-11-02 · Original record ↗
- arxiv
- 2610.40947
Explicit source versions (1)
- v1 · Date not supplied
- Preprint available
Source date: 2026-11-02 · First detected: 2026-11-02 UTC · Historical baseline
Source evidence (1)
- Official evidence ↗ · RSS guid=oai:arXiv.org:2610.40947v1; arxiv:announce_type=new; item pubDate (official metadata)
Abstract · arxiv
Real-time 4D occupancy forecasting requires reliably extrapolating both ego motion and surrounding actor motion under strict latency budgets. Recent occupancy world models improve long-horizon fidelity through generative decoders and 3D volumetric modeling, yet their computational cost makes real-time deployment challenging. Light-weight BEV forecasters meet latency requirements but often blur or suppress fast-moving actors at longer horizons. We present UFMF (Ultra-Fast Multi-modal Forecaster), a real-time forecaster that keeps an attention-free diagonal state-space temporal core while injecting targeted kinematic and geometric structure where lightweight models typically fail. A radar-guided velocity-aware dynamic prior (VADP) decomposes future displacement into a linear extrapolation plus a learned horizon-dependent residual. The resulting per-cell flow fields transport a compact dynamic latent subspace during rollout. Two auxiliary training constraints complement this branch: world-frame dynamic supervision enforces inter-horizon trajectory coherence, and class-conditioned vertical-profile supervision regularizes BEV-to-3D lifting. UFMF operates at 122 FPS on a single RTX 4090 and reaches 0.45 fast-displacement mIoU on nuScenes, outperforming recent occupancy world models at a fraction of their computational cost.
