FreeOcc improves 4D occupancy forecasts by removing self-motion effects
FreeOcc: Decoupling Ego-Motion for Efficient 4D Occupancy Forecasting via Continuous Flow Matching
Computer Vision and Pattern Recognition
Summary
Understanding how the world around a self-driving car changes over time is tricky because the car itself is moving. The authors created FreeOcc, a new way to separate the car’s own movements from other changes in the scene. This clever separation lets their model work faster and use less computing power. FreeOcc also focuses more on uncertain areas to improve predictions, and it performs well even when sensors lose some data or timing is off.
What this means in practice
- •For autonomous vehicle engineers: Build more efficient and accurate models for predicting dynamic environments around self-driving cars by explicitly removing the cars’ own motion effects.$Commercial implications: Enables advanced real-time scene prediction modules sold to autonomous vehicle manufacturers to improve safety and efficiency.
- •For robotics perception teams: Deploy computationally lightweight world models that maintain robustness to sensor noise and timing errors for robots operating in complex environments.
Authors
Zeping Zhang, Zhuoya Zhao, Samy Metari, Robert Laganière
Abstract
Current 4D occupancy world models use heavy cross-attention or deep recurrence to disentangle ego-motion from scene dynamics. This requires significant computational capacity. However, this effort is largely redundant since the ego-vehicle's intended future motion is provided by the downstream motion planner. We present FreeOcc, a lightweight continuous-time world model that introduces explicit $SE(2)$ geometric decoupling into a conditional flow-matching framework. By mathematically factoring out self-motion, we create a purely dynamic feature space. This space supports attention-free temporal fusion and allows physics-constrained ODEs to operate strictly on residual agent kinematics. To maximize deployment efficiency, FreeOcc introduces an uncertainty-aware blending mechanism. This routes well-observed voxels to a single-pass decoder, reserving the ODE solver exclusively for ambiguous, high-uncertainty regions. Requiring only 28.3M parameters-roughly 40% of comparable models-our optimal ensemble mode (FreeOcc-E) runs at 27 FPS. On nuScenes, FreeOcc achieves state-of-the-art forecasting performance with a 0.47 m Near Field Chamfer Distance (NFCD) and 32.1% mIoU. Furthermore, the model transfers zero-shot to SemanticKITTI. It also proves highly robust, retaining 92.8% of its ideal mIoU under significant sensor packet loss and timestamp jitter.