AMP reframes robot manipulation as pixel classification to dodge action-space explosion
jiqizhixin · x · 2026-09-18
LLMs showed cross-entropy prediction scales well on 1D sequences, but robot actions are high-dimensional: 6–7 DOF per timestep, so even coarse discretization (10 bins per dimension) yields 1 million possible joint actions, and finer discretization or longer horizons make the space explode — so the cross-entropy paradigm behind LLMs can't directly extend to robot learning.
Action Map Policy (AMP) proposes a new paradigm for robot manipulation learning: it reframes high-precision 3D closed-loop manipulation as pixel classification in image space. Instead of predicting action tokens in a combinatorially exploding space, the model classifies where to act on an action map — a 2D representation encoding 3D motion as spatial positions — bringing cross-entropy prediction from language into robotics without the dimensionality blowup.
More from Embodied
- LAIA dataset: 15 hours of CARLA driving with human gaze labels for explainable end-to-end AV research — abursuc · 2026-09-18
- Vintage robotics video: a handful of PhDs testing modular architectures in the wild — abursuc · 2026-09-18
- Robotics startup founder: hosting events self-selects the hard-to-find ML talent — DominiqueCAPaul · 2026-09-18
- Developers embrace open-source physics simulator MuJoCo for modular robot prototyping — Obijuan_cube · 2026-09-18
- Stanford's FAMOS infers 3D articulation from sparse partial point clouds feed-forward — Stanford-University · 2026-09-18
- Waymo to launch autonomous robotaxis in Singapore by 2028, its first Southeast Asia market — emmanuelvivier · 2026-09-18