RL Zero combines video generation models with unsupervised RL to generate zero-shot prompt-to-policy behavior. This can help generate desired behaviors without complex reward design.
Check out this exciting work at
- arxiv.org/abs/2412.05718
- hari-sikchi.github.io/rlzero/
arxiv.org
RL Zero: Zero-Shot Language to Behaviors without any Supervision
Rewards remain an uninterpretable way to specify tasks for Reinforcement Learning, as humans are often unable to predict the optimal behavior of any given reward function, leading to poor reward desig...