The Perception & Action Podcast 552 – Turvey, Lectures on Perception: An Ecological Perspective, Chapters 17-19 (JC54)
Oct 14, 2025
Join cognitive scientist Andrew Wilson, who dives deep into perception and ecological theory, and researcher Marianne Davies, who connects these ideas to coaching and robotics. They explore why categorization often fails in understanding action and how optic flow offers better navigation cues than object recognition. The discussion reveals the limitations of symbolic AI, emphasizing embodied robotics. Furthermore, they challenge traditional views of language in perception, advocating for direct perception as a more viable approach.
AI Snips
Chapters
Books
Transcript
Episode notes
Spider Robot Leg Failure
- Marianne describes two spider-robot controllers where the exploratory one adapted after leg damage while the tuned controller failed.
- The exploratory robot found robust solutions; hard-coded optimization proved brittle.
Embodiment Generates Behavior
- Passive dynamic walkers demonstrate that embodiment and mechanics can generate biological movement without control computation.
- Physical design and constraints can produce robust behavior independent of symbolic processing.
Affordances Beat Labels For Action
- Agents (animals, humans) avoid collisions by perceiving affordances like looming and time-to-contact, not by labeling objects.
- Rob and Andrew emphasize that recognizing 'car' adds nothing to action control compared with optic-flow cues like tau.
