Manipulation as in Simulation: Enabling Accurate Geometry Perception in Robots
Modern robotic manipulation primarily relies on visual observations in a 2D color space for skill learning but suffers from poor generalization. In contrast, humans, living in a 3D world, depend more on physical properties-such as distance, size, and shape-than on texture when interacting with objects.
Evidence notes
- The paper studies geometry-aware perception for robotic manipulation beyond conventional 2D colour observations.
- Physical experiments use ZED 2i to capture distance, size and shape information.
- The evaluation tests whether geometry-focused inputs improve visuomotor-policy generalisation.
Company context
Stereolabs develops ZED stereo cameras, depth perception hardware and spatial AI software for robots, drones and autonomous systems.