NeurIPS 2019
Sun Dec 8th through Sat the 14th, 2019 at Vancouver Convention Center
Paper ID:2016
Title:MCP: Learning Composable Hierarchical Control with Multiplicative Compositional Policies


		
This paper proposes an interesting method that learns multiple composable policies that can produce a continuous spectrum of diverse actions, where primitives can be transferred to new tasks. The paper is well written, and the experimental results are convincing. The ablation study would strengthen the paper.