Growing Q-Networks: Solving Continuous Control Tasks with Adaptive Control Resolution
This addresses the need for smooth control in robotics to reduce wear and energy use, while maintaining exploration benefits, though it is incremental as it builds on existing decoupled Q-learning methods.
The paper tackles the performance gap between coarse action discretizations and smooth control signals in continuous control tasks by growing discrete action spaces from coarse to fine resolution, achieving strong performance on tasks with up to 38-dimensional action spaces.
Recent reinforcement learning approaches have shown surprisingly strong capabilities of bang-bang policies for solving continuous control benchmarks. The underlying coarse action space discretizations often yield favourable exploration characteristics while final performance does not visibly suffer in the absence of action penalization in line with optimal control theory. In robotics applications, smooth control signals are commonly preferred to reduce system wear and energy efficiency, but action costs can be detrimental to exploration during early training. In this work, we aim to bridge this performance gap by growing discrete action spaces from coarse to fine control resolution, taking advantage of recent results in decoupled Q-learning to scale our approach to high-dimensional action spaces up to dim(A) = 38. Our work indicates that an adaptive control resolution in combination with value decomposition yields simple critic-only algorithms that yield surprisingly strong performance on continuous control tasks.