PokeBot's RoboAPI brings zero-shot, natural-language robot control to any compatible arm
haider1 · x · 2026-10-30
PokeBot launched RoboAPI, a public API that takes a robot's observations and state and predicts the next action chunk in near real time from natural-language instructions — no task-specific training. The key is a UMI-inspired post-training configuration: a wrist-mounted fisheye camera plus end-effector pose control. Results: swapping fisheye brands (Insta360 Ace Pro2 → DJI Action 6) barely affects performance, and the model transferred to entirely unfamiliar setups including UR7e → YAM Standard. The upshot: any arm with capable inverse kinematics just needs an action camera on the wrist for RoboAPI to control it — an attempt to bring the GPT-3-style foundation-model-API model to the physical world.
More from Embodied
- Standard Bots doc goes live: 7 years to sell its first 100 robots — Rewkang · 2026-10-30
- Neuralink ships 8 brain-interface features in 72 hours, lets ALS patient order coffee by thought — XFreeze · 2026-10-30
- GitHub dev advocate shows off e-ink conference badge displaying live follower and repo stats — 0xkarasy · 2026-10-30
- Neuralink turns an ALS patient's thoughts into speech as he orders coffee via Bloom app — XFreeze · 2026-10-30
- Generalist AI ships GEN-1.5 and challenges the community to SLAM its wrist-cam data — E0M · 2026-10-30
- VisualForce: robots infer grip force from soft gripper deformation, no force sensors needed — du_yilun · 2026-10-30