Event
CHOP: Counterfactual Human Preference Labels Improve Navigation Obstacle Avoidance
1 reports · 1 sources
Gershom Seneviratne and co-authors propose the CHOP method, which fine-tunes visuomotor navigation policies using pairwise preference labels on trajectories from human annotators (based on expected outcomes such as collision risk and path efficiency), aligning them with human intuitions of safety and obstacle avoidance. On the SCAND dataset, after CHOP fine-tuning, multiple advanced models reduced near-collision events by an average of 49.7% compared with their respective pretrained baselines, reduced deviations from human-preferred trajectories by 45.0%, and increased average obstacle clearance by 19.8%. The authors state that these improvements also transferred to real-world deployment on the Ghost Robotics Vision60 quadruped robot: compared with the pretrained baseline, average target success rate increased by 24.4%, minimum obstacle clearance increased by 6.8%, collision and manual intervention events decreased by 45.7%, and normalized path completion improved by 38.6%.
Report timeline
- CHOP: Counterfactual human preference labels improve obstacle avoidance in visuomotor navigation policies
The authors propose CHOP, which fine-tunes visuomotor navigation policies with counterfactual human preference labels to align them with human intuition about safety and obstacle avoidance. On the SCAND dataset, multiple advanced models fine-tuned with CHOP reduced near-collision events by an average of 49.7%, decreased deviation from human-preferred trajectories by 45.0%, and increased average obstacle clearance by 19.8%.
arXiv cs.RO ·
Related topics
Interest trend
- Current
- 3
- Peak
- 4Nov 4 01:00
- Past 24 hours
- –
The trend compares participants observed throughout the same period, so its range may be smaller than the current interest count. Move or tap to inspect each hour; use the arrow keys on a keyboard.