Belief-Space Perception Routing under Coupled Sensor Faults and Compute Contention
A perception router tracks probabilistic estimates of sensor-fault and compute-contention states, couples them with a noisy-OR term, and selects among four YOLO detector configurations to meet frame deadlines. Under co-occurring stressors, the coupled policy reduces deadline-miss rate by 1.1 to 9.4 percentage points versus independent treatment, with sign-test p=0.001. Routing costs tens of microseconds per frame.
发展脉络
- 首次出现Belief-Space Perception Routing under Coupled Sensor Faults and Compute ContentionarXiv cs.RO
- 当前判断This research highlights the importance of holistic resource management in autonomous systems, where perception, planning, and control share limited compute. It suggests that treating perception as an isolated component is suboptimal. The approach could influence the design of next-generation robotic platforms, emphasizing the need for adaptive perception pipelines that respond to both environmental and computational conditions.Agent Pulse · 分析
A robot perception system faces simultaneous challenges: camera degradation in adverse weather and compute contention from shared onboard processing. The proposed perception router maintains probabilistic estimates of both sensor-fault and compute-contention states, coupling them via a noisy-OR term to select among four YOLO detector configurations (YOLO11x/n at 1280 or 640 px) to meet frame deadlines. Experiments show that when both stressors co-occur, the coupled policy reduces deadline-miss rate by 1.1 to 9.4 percentage points compared to treating them independently, with statistical significance (sign-test p=0.001). The routing overhead is tens of microseconds per frame. The study also investigates whether coupling arises naturally in real data, using eight RADIATE adverse-weather sequences.
The method demonstrates that coupling sensor-fault and compute-contention estimates improves deadline adherence in perception systems. The noisy-OR coupling term allows the router to account for correlated stressors, leading to better detector selection. The routing overhead is minimal, making it practical for real-time systems. Future work could explore adaptive coupling or online learning of the coupling parameters.
This research highlights the importance of holistic resource management in autonomous systems, where perception, planning, and control share limited compute. It suggests that treating perception as an isolated component is suboptimal. The approach could influence the design of next-generation robotic platforms, emphasizing the need for adaptive perception pipelines that respond to both environmental and computational conditions.
For companies developing autonomous vehicles or robots, this research offers a way to improve reliability in challenging conditions without additional hardware. By reducing deadline misses, it can enhance safety and performance, potentially reducing development costs and time-to-market for robust perception systems.
The next signal to watch is whether the coupled perception routing approach is adopted in real-world autonomous systems, particularly in adverse weather conditions. Also, whether the method scales to more complex perception stacks and whether it can be integrated with other resource management techniques.