Research map / Reinforcement learning / Visual and representation RL
Evaluating Model-Based Planning and Planner Amortization for Continuous Control
Arunkumar Byravan, Leonard Hasenclever, Piotr Trochim, Mehdi Mirza, Alessandro Davide Ialongo, Yuval Tassa, Jost Tobias Springenberg, Abbas Abdolmaleki, Nicolas Heess, Josh Merel, Martin Riedmiller
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 changed | Where | Measured | Effect |
|---|---|---|---|
| Use a pre-trained task-agnostic proposal instead of a zero-mean Gaussian proposal for MPC | MPC with learned or ground-truth dynamics · Ant and OP3 locomotion tasks | reward: 56± 1 -> 420±14 (Ant GTTP); 36± 1 -> 178±4 (OP3 GTTP) | Improved |
| Combine MPC data collection with MPO policy learning | MPC+MPO · Ant and OP3 GTTP | actor performance | Improved |
| Add behavioral cloning of planner-generated data to MPO | MPO · GTTP tasks | proposal performance | Improved |
| Combine MPC data collection with MPO and behavioral cloning | MPC+MPO+BC · GTTP tasks | learning speed | Improved |
| Use MPC data collection with MPO without behavioral cloning | MPC+MPO · GTTP tasks | proposal performance | No clear effect |
| Add behavioral cloning to MPC+MPO | MPC+MPO · Ant and OP3 walking | performance | No clear effect |
| Transfer a learned model to the target task, either frozen or finetuned | MPC+MPO+BC · OP3 GTTP and cross-task transfers | learning speed | No clear effect |
| Keep the transferred model frozen rather than finetuning it | MPC+MPO+BC · OP3 GTTP | performance | Hurt |
| Transfer a proposal from forward walking to GTTP or from GTTP to backward walking | MPO+BC and MPC+MPO+BC · Cross-task transfer | learning speed | Hurt |
| Always plan rather than interleave planning and proposal actions (p_plan=1.0 instead of p_plan=0.5) (ablation) | MPC+MPO+BC · OP3 GTTP | performance | Hurt |
| Increase planner sample count above the default N=250 (ablation) | MPC+MPO+BC · OP3 GTTP | performance | No clear effect |
| Use a longer planning horizon, H=20, instead of the default H=10 (ablation) | MPC+MPO+BC · OP3 GTTP | performance | No clear effect |
Closest papers by meaning
- Bootstrapped Model Predictive Control (ICLR 2025 · similarity 0.90)
- Exploring Model-based Planning with Policy Networks (ICLR 2020 · similarity 0.89)
- Learning from Reward-Free Offline Data: A Case for Planning with Latent Dynamics Models (NeurIPS 2025 · similarity 0.86)
- Improving planning and MBRL with temporally-extended actions (NeurIPS 2025 · similarity 0.85)
- World Models as Reference Trajectories for Rapid Motor Adaptation (NeurIPS 2025 · similarity 0.85)
- Overcoming Slow Decision Frequencies in Continuous Control: Model-Based Sequence Reinforcement Learning for Model-Free Control (ICLR 2025 · similarity 0.85)
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.
