Reproduce “Experience-Based Feasibility-Aware Generative Adversarial Imitation from Observation under Embodiment Mismatch” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Yoshiki Takebayashi, Giovanni Perantoni, Hikaru Sasaki, Matteo Saveriano, Takamitsu Matsubara
EF-GAIfO learns to weight state-only demonstrations by comparing their transitions with transitions collected through the robot’s policy experience. A reconstruction model estimates similarity, and its reference broadens as experience grows. In point-mass tests, it outperformed GAIfO and WGAIfO; on a Unitree Go2, it achieved 80% success on random-target grasping and lifting in ten trials, a result limited to that task.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.