MobileVISTA: Pose Generalization from Single-Pose Demos
Loading video
Loading videoMobile manipulation policies trained on demos from a single base pose break when deployment pose drifts by a few centimeters. MobileVISTA turns those canonical-pose demos into pose-perturbed training data by jointly augmenting egocentric observations and retargeting actions for the base offset, needing no extra demos and no trained generative model; validated in simulation (humanoid and bimanual) and on a real Galaxea R1 Pro.
Category: humanoid
Author: @s_wistreich
Date: 2026-10-07T00:00:00
Duration: 15.0s
Reference: https://arxiv.org/abs/2610.07511
