3D Diffusion Policy: Generalizable Visuomotor Policy Learning via Simple 3D Representations

Imitation learning provides an efficient way to teach robots dexterousskills; however, learning complex skills robustly and generalizablely usuallyconsumes large amounts of human demonstrations. To tackle this challengingproblem, we present 3D Diffusion Policy (DP3), a novel visual imitationlearning approach that incorporates the power of 3D visual representations intodiffusion policies, a class of conditional action generative models. The coredesign of DP3 is the utilization of a compact 3D visual representation,extracted from sparse point clouds with an efficient point encoder. In ourexperiments involving 72 simulation tasks, DP3 successfully handles most taskswith just 10 demonstrations and surpasses baselines with a 24.2

Further reading#

• Access Paper in arXiv.org