Diminishing Return of Value Expansion Methods
This challenges the common assumption that model accuracy is the primary bottleneck in model-based reinforcement learning, which is significant for researchers and practitioners in the field as it suggests focusing efforts elsewhere for efficiency gains.
The paper investigates the sample efficiency gains from improved dynamics models in model-based reinforcement learning, finding that even perfect models yield diminishing returns and only marginal improvements compared to learned models, with model-free methods achieving comparable performance without computational overhead.
Model-based reinforcement learning aims to increase sample efficiency, but the accuracy of dynamics models and the resulting compounding errors are often seen as key limitations. This paper empirically investigates potential sample efficiency gains from improved dynamics models in model-based value expansion methods. Our study reveals two key findings when using oracle dynamics models to eliminate compounding errors. First, longer rollout horizons enhance sample efficiency, but the improvements quickly diminish with each additional expansion step. Second, increased model accuracy only marginally improves sample efficiency compared to learned models with identical horizons. These diminishing returns in sample efficiency are particularly noteworthy when compared to model-free value expansion methods. These model-free algorithms achieve comparable performance without the computational overhead. Our results suggest that the limitation of model-based value expansion methods cannot be attributed to model accuracy. Although higher accuracy is beneficial, even perfect models do not provide unrivaled sample efficiency. Therefore, the bottleneck exists elsewhere. These results challenge the common assumption that model accuracy is the primary constraint in model-based reinforcement learning.