Timezone: »

Learning to Constrain Policy Optimization with Virtual Trust Region
Thai Hung Le · Thommen Karimpanal George · Majid Abdolshah · Dung Nguyen · Kien Do · Sunil Gupta · Svetha Venkatesh


We introduce a constrained optimization method for policy gradient reinforcement learning, which uses two trust regions to regulate each policy update. In addition to using the proximity of one single old policy as the first trust region as done by prior works, we propose forming a second trust region by constructing another virtual policy that represents a wide range of past policies. We then enforce the new policy to stay closer to the virtual policy, which is beneficial if the old policy performs poorly. We propose a mechanism to automatically build the virtual policy from a memory buffer of past policies, providing a new capability for dynamically selecting appropriate trust regions during the optimization process. Our proposed method, dubbed Memory-Constrained Policy Optimization (MCPO), is examined in diverse environments, including robotic locomotion control, navigation with sparse rewards and Atari games, consistently demonstrating competitive performance against recent on-policy constrained policy gradient methods.

Author Information

Thai Hung Le (Deakin University)
Thommen Karimpanal George (Deakin University)
Majid Abdolshah (Amazon)
Dung Nguyen (Deakin University)
Kien Do (Deakin University)
Sunil Gupta (Deakin University)
Svetha Venkatesh (Deakin University)

Related Events (a corresponding poster, oral, or spotlight)

More from the Same Authors