Get the latest tech news

Show HN: Shoggoth Mini – A soft tentacle robot powered by GPT-4o and RL


Matthieu Le Cauchois's website

To address this, I added a breathing idle mode with small, noisy oscillations that shift between principal directions, keeping it feeling alive even when not actively responding. Note that you could use curriculum learning to gradually transition from this 2D constraint to full 3D control by starting with the simplified representation and progressively expanding the action space as the policy becomes more stable. Ditching the constrained 2D control space Expanding the expression repertoire, both open and closed-loop, potentially through RLHF Adding more tentacles and teaching it to crawl Using direct drive motors to reduce noise

Get the Android app

Or read this on Hacker News

Read more on:

Photo of GPT-4o

GPT-4o

Photo of Shoggoth Mini

Shoggoth Mini

Photo of soft tentacle robot

soft tentacle robot

Related news:

News photo

GPT-4o shows humanlike patterns of cognitive dissonance moderated by free choice

News photo

GPT-4o draws itself as a consistent type of guy

News photo

Contextual AI’s new AI model crushes GPT-4o in accuracy — here’s why it matters