Posterior Behavioral Cloning: Pretraining BC Policies for Efficient RL Finetuning

Standard practice across domains from robotics to language is to first pretrain a policy on a large-scale demonstration dataset, and then finetune this policy, typically with reinforcement learning (RL), in order to improve performance on deployment domains. This finetuning step has proved critical in achieving human or super-human performance, yet while much attention has been given to developing more effective finetuning algorithms, little attention has been given to ensuring the pretrained policy is an effective initialization for RL finetuning. In this work we seek to understand how the pretrained policy affects finetuning performance, and how to pretrain policies in order to ensure they are effective initializations for finetuning. We first show theoretically that standard behavioral cloning (BC) can fail to ensure coverage over the demonstrator's actions, a minimal condition necessary for effective RL finetuning. We then show that if, instead of exactly fitting the observed demonstrations, we train a policy to model the posterior distribution of the demonstrator's behavior given the demonstration dataset, we do obtain a policy that ensures coverage over the demonstrator's actions, enabling more effective finetuning. Furthermore, this policy achieves this while ensuring pretrained performance is no worse than that of the BC policy. We then show this approach is practically implementable with modern generative models and leads to significantly improved RL finetuning performance on both realistic robotic control benchmarks and real-world robotic manipulation tasks, as compared to standard behavioral cloning.

We introduce an approach for training policies from demonstrations that ensures they are an effective initialization for online improvement. In particular, we show that standard behavioral cloning can overfit to the demonstrations, leading to poor expressivity, while by instead fitting a distribution that takes into account our uncertainty of the demonstrator's behavior we can match the performance of the behavioral cloning policy, while achieving greater expressivity. We show that this leads to improved performance in RL finetuning without damaging performance of the policy trained from demonstrations.