AI

EyeRobot 2.0: Active Gaze for Precise Manipulation without Wrist Cameras

Read full article at source ↗

Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 05/10/2026, 19:18:24 UTC.

Briefings

EyeRobot 2.0 enables precise bimanual manipulation using only a single stereo camera by employing active gaze to attend to a 3D fixation point. The system uses hierarchical reinforcement learning to train a low-level gaze servoing policy and a target selector that emits fixation goals based on task progress. Teleoperation data was collected for 7 real-world and 6 simulated tasks, with over 1000 physical and 1800 simulated trials conducted comparing EyeRobot 2.0 against passive stereo and ego + wrist camera policies. Removing wrist cameras significantly impacts performance, reducing real-world success from 52% to 27% with passive stereo, whereas EyeRobot 2.0 maintains high success rates even when grasped objects occlude wrist cameras.

Key points

• EyeRobot 2.0 uses active gaze to physically attend to a 3D fixation point, swiveling two eye viewpoints to center their gaze on the target.
• The system achieves 40% higher success rates in real-world tasks compared to passive stereo when wrist cameras are removed, matching ego + wrist policies when their views are clear.

Context and limitations

Data collection was limited to 7 real-world and 6 simulated tasks, with over 1000 physical and 1800 simulated trials conducted. Results are based on teleoperation data and may not generalize to all scenarios.

Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 14/10/2026, 15:49:14 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.

View archived briefing on IPFS
bafybeihnpas4rq7ruwjpxfwnsfekjqdiqoq572ho3w3y6pywycli53zwze