Reproduce “Reframing Object-Centric Reinforcement Learning: From Absolute to Relative Representations” 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
Cedric Derstroff, Jannis Blüml, Janosch Moos, Michael Girstl, Zihan Ye, Debora Clever, Kristian Kersting
ORCA gives object-centric reinforcement-learning policies positions and velocities relative to an anchor, so shifted scenes can produce identical inputs. ORBiT processes these tokens with a transformer; ORCA adds no learnable parameters but requires a localizable anchor. Across 13 Atari games, ORBiT raises HNS IQM from 0.54 to 1.10 versus absolute-input OCT, though gains depend on anchor choice.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.