Embodied AI Glossary中文

Exteroception

外部感知Advanced

A robot’s sensing of its surrounding environment using sensors like cameras and lidar.

Exteroception comes from physiology, referring to an organism’s ability to sense stimuli from outside the body — sight, hearing, skin touch — as opposed to proprioception, which senses the body’s own limb position and motion. Robotics keeps this same distinction: cameras, depth cameras, lidar, ultrasonic sensors, and tactile sensors that measure the environment fall under exteroception, while joint encoders, IMUs, and motor current that measure the robot’s own state fall under proprioception. Exteroception lets a robot see steps, obstacles, and target objects ahead of time, but it is vulnerable to lighting, occlusion, reflections, and noise; proprioception is stable, but it can only sense terrain once contact has already happened. Legged robots therefore often fuse the two: adjusting gait ahead of time when exteroceptive information is reliable, and falling back on proprioception when it isn’t.

ExampleETH Zurich’s legged locomotion work (Science Robotics 2022) end-to-end fuses exteroception and proprioception with an attention-based recurrent encoder, letting the robot adjust its gait before making contact with terrain, and it completed an Alpine hiking trail within the time recommended for human hikers.

Also called
Exteroceptive Sensing, Environment Perception
Related
Proprioception · Multi-Sensor Fusion · Perceptive Locomotion · Blind Locomotion · State Estimation · LiDAR
Sources
Wikipedia: Exteroception
Learning robust perceptive locomotion for quadrupedal robots in the wild (Science Robotics 2022)

See it in the full glossary →