Reproduce “Temporal Behavior Trees for Reinforcement Learning: Specifications as Rewards, Curricula, and Diagnostics” 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
Vladimir Krsmanovic, Bernd Finkbeiner
TBTCL uses one Temporal Behavior Tree to specify tasks, shape reinforcement-learning rewards, derive curricula, and diagnose behavior. Its Signal Temporal Logic leaves track robustness changes, with added safety penalties and parallel-progress feedback. On PointNav reach-avoid, it achieves mean satisfaction of 0.64 ± 0.09 versus 0.00 ± 0.00 for STL Shaping; sequence and repetition results match, across six specifications in two small domains.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.