FoMo Rewards: Can we cast foundation models as reward functions?
This work opens possibilities for designing open-ended agents in interactive tasks, but it appears incremental as it builds on existing foundation models without major methodological breakthroughs.
The authors tackled the problem of using foundation models as generic reward functions for reinforcement learning by proposing a pipeline that combines a vision model with a large language model to infer task likelihoods from observations, showing it associates high values with desired behaviors and lower values for incorrect policies.
We explore the viability of casting foundation models as generic reward functions for reinforcement learning. To this end, we propose a simple pipeline that interfaces an off-the-shelf vision model with a large language model. Specifically, given a trajectory of observations, we infer the likelihood of an instruction describing the task that the user wants an agent to perform. We show that this generic likelihood function exhibits the characteristics ideally expected from a reward function: it associates high values with the desired behaviour and lower values for several similar, but incorrect policies. Overall, our work opens the possibility of designing open-ended agents for interactive tasks via foundation models.