Agentic robotics keeps bolting GPT-6 onto harnesses — but where are systems that learn from physical failure?
DJiafei · x · 2026-09-25
The author notes a wave of "agentic robotics" work that is essentially GPT-6 Astra plus a robotics harness, and asks what's new beyond the backbone. VoxPoser, Code as Policies, and Manipulate-Anything already explored LLM orchestration, with deeper roots in TAMP research.
- Stronger models improve results, but "swap in a better model" is not itself the research question
- The real question: where are systems that learn from physical failure, improve through interaction, and need less hand-engineered scaffolding?
A pointed methodological critique of the current embodied-AI paradigm.
More from Embodied
- Musk on China's humanoid robot games: the future will have 'a lot, a lot' of robots — XFreeze · 2026-09-25
- Meta sidesteps the VR/XR/MR jargon mess by just calling them VR glasses — pvncher · 2026-09-25
- Magnetic-footed quadruped climbs vertical ship hulls to weld and inspect — TinfoilTricorn · 2026-09-25
- Robotics researcher: scaling atoms is the neglected half, Western labs bet on the East — yunta_tsai · 2026-09-25
- Hobbyist trains Asimov robot to do a 'Ranger Walk' after 10 months of simulated walking on A100s — freelerobot · 2026-09-25
- DIY soft pouch actuator gets 4x stroke via TPU-silicone hybrid design — IanPritchard · 2026-09-25