Amazon FAR's FIBER uses action-aligned latent planning to let Unitree G1 open doors and carry boxes
ChongZzZhang · x · 2026-10-21
- Amazon FAR with CMU and Stanford introduce FIBER, a framework for humanoid loco-manipulation that plans in an action-aligned latent space.
- A planner outputs latent commands to a frozen, pretrained Behavior Foundation Model (BFM); each latent is defined by the actions it induces, so the BFM closes the loop at 50 Hz while the planner replans at 5 Hz.
- Trained entirely offline in simulation from heterogeneous experts, one planner performs heavy-door pushing, box carrying (including thrown boxes), object grasping, and controllable navigation on a real Unitree G1 using onboard depth.
- Paper on arXiv; code coming soon, live demo on the project site.
More from Embodied
- Robotics startup launches Deployments to serve fine-tuned HF checkpoints to robots in real time — arnie_hacker · 2026-10-21
- Sentdex trains robots to go from lying down to a half kneel, eyeing a full stand — Sentdex · 2026-10-21
- NVIDIA open-sources MotionBricks: 350K motion skills at 15,000 FPS, 2ms latency — aftahi_ai · 2026-10-21
- NVIDIA Cosmos3-H-Dreams: Generative Simulation Brings Clinical Surgery to Life — Hugging Face Blog · 2026-10-21
- FLoTS-SLAM uses optical-flow residuals to hit 0.73 DM-IoU in dynamic monocular SLAM — zhenjun_zhao · 2026-10-21
- Plato ring launches at $199 to turn near-silent speech into AI commands — TinfoilTricorn · 2026-10-21