AI Paper Atlas

Research map / Reinforcement learning / Visual and representation RL

MCP: Learning Composable Hierarchical Control with Multiplicative Compositional Policies

Xue Bin Peng, Michael Chang, Grace Zhang, Pieter Abbeel, Sergey Levine

NeurIPS 2019 · Reinforcement learning › Visual and representation RL · approach: Curriculum and environment design · 54 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
Use MCP instead of the strongest listed baseline, FinetuneBiped heading control · Heading: Bipednormalized return: 0.970 \pm 0.002 -> 0.976 \pm 0.002Improved
Use MCP instead of FinetuneHumanoid heading control · Heading: Humanoidnormalized return: 0.975 \pm 0.008 -> 0.970 \pm 0.003Hurt
Use MCP instead of FinetuneT-Rex heading control · Heading: T-Rexnormalized return: 0.953 \pm 0.004 -> 0.932 \pm 0.007Hurt
Use multiplicative MCP composition instead of the latent-space modelBiped box carrying · Carry: Bipednormalized return: 0.456 \pm 0.031 -> 0.575 \pm 0.032Improved
Use multiplicative MCP composition instead of the latent-space modelBiped ball dribbling · Dribble: Bipednormalized return: 0.768 \pm 0.012 -> 0.782 \pm 0.008Improved
Use multiplicative MCP composition instead of the latent-space modelHumanoid ball dribbling · Dribble: Humanoidnormalized return: 0.751 \pm 0.006 -> 0.805 \pm 0.006Improved
Use multiplicative MCP composition instead of the latent-space modelT-Rex ball dribbling · Dribble: T-Rexnormalized return: 0.115 \pm 0.013 -> 0.781 \pm 0.021Improved
Use MCP instead of ScratchAnt holdout heading control · Holdout: Antnormalized return: 0.951 \pm 0.093 -> 0.812 \pm 0.030Hurt
Increase primitive count from 4 to 8 (ablation)MCP transfer policies · Transfer taskslearning performanceNo clear effect
Increase primitive count beyond 8, including to 32 (ablation)MCP transfer policies · Transfer taskslearning efficiencyHurt
Fine-tune the intra-option policies and termination functions during transfer (ablation)Option-Critic · Transfer tasksperformanceNo clear effect

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy