notesum.ai

Published at October 22

Composing Diffusion Policies for Few-shot Learning of Movement Trajectories

cs.RO

cs.AI

Released Date: October 22, 2024

Authors: Omkar Patil¹, Anant Sah¹, Nakul Gopalan¹

Aff.: ¹Unknown

Arxiv: https://arxiv.org/abs/2410.17479v1

Refer to caption

	Number	Vanilla	Fine-tuned	Diffusion
	of	Composition	Policy	Score
Trajectories	demos			Equilibrium
Spiral X	5	0.58	0.64	0.51
	15	0.58	0.26	0.09
	40	0.58	0.15	0.09
Step	5	0.79	0.50	0.25
	15	0.18	0.27	0.20
	40	0.15	0.17	0.12
OSC X + Line XZ	5	0.75	0.57	0.32
	15	0.30	0.25	0.06
	40	0.37	0.14	0.12