Learning Bed-to-Chair Transfer
with Whole-Body RL

Elle Miller

University of Edinburgh

Ball Environment

Five random seeds. Left: RL baseline. Right: our method.

Human (Near)

Five random seeds. Left: RL baseline. Right: our method.

Human (Far)

Five random seeds. Left: RL baseline. Right: our method.

Random Force Perturbation (300-500N)

Ball

Human

Mass Generalisation (11-33 kg, one policy)

Human mass colourbar, viridis, 11 to 33 kilograms

Size Generalisation

Height 116cm (3.8 ft)

Height 178cm (5.8 ft)

Ablation Study

SIM-only, no interaction gate or hinted object path

SIM with interaction gate, no hinted object path