Imagination at Inference: Synthesizing In-Hand Views for Robust Visuomotor Policy Inference
Visual observations from different viewpoints can significantly influence the performance of visuomotor policies in robotic manipulation. Among these, egocentric (in-hand) views often provide crucial information for precise control.
Evidence notes
- The paper synthesises in-hand views at inference time to improve visuomotor manipulation policies.
- Physical experiments use ZED 2i as the external perception platform.
- The evaluation studies robustness when direct egocentric observations are limited or unavailable.
Company context
Stereolabs develops ZED stereo cameras, depth perception hardware and spatial AI software for robots, drones and autonomous systems.