Research map / Reinforcement learning / Visual and representation RL
Myriad: a real-world testbed to bridge trajectory optimization and deep learning
Nikolaus H. R. Howe, Simon Dufort-Labbé, Nitarshan Rajkumar, Pierre-Luc Bacon
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 changed | Where | Measured | Effect |
|---|---|---|---|
| train end-to-end through unrolled NLP solver instead of sysid then planning on Cart-Pole | Neural ODE model · Cart-Pole Swing-Up | cost: 242.25 -> 36.90 | Improved |
| end-to-end training instead of Neural ODE sysid on Pendulum | Neural ODE model · Pendulum | cost: 20.88 -> 1.90 | Improved |
| end-to-end training instead of Neural ODE sysid on Predator Prey | Neural ODE model · Predator Prey | cost: 1.95 -> 1.04 | Improved |
| end-to-end training instead of Neural ODE sysid on Van Der Pol | Neural ODE model · Van Der Pol | cost: 11.12 -> 16.81 | Hurt |
| end-to-end training instead of Neural ODE sysid on Mountain Car | Neural ODE model · Mountain Car | cost: 15.87 -> 3000 | Hurt |
| end-to-end training instead of Neural ODE sysid on Timber Harvest | Neural ODE model · Timber Harvest | cost: -3928.77 -> -500.00 | Hurt |
| end-to-end training instead of Neural ODE sysid on Bacteria | Neural ODE model · Bacteria | cost: -2.72 -> -7.60 | Improved |
| end-to-end training instead of Neural ODE sysid on Tumour | Neural ODE model · Tumour | cost: 8468.68 -> 9161.23 | Hurt |
| learn structured parametric model instead of using true parameters | trajectory optimization · Pendulum | cost: 25.75 -> 25.49 | No clear effect |
| learn Neural ODE instead of parametric model for dynamics | trajectory optimization · Cart-Pole Swing-Up | cost: 87.78 -> 242.25 | Hurt |
| smaller training set of 10 trajectories instead of 100 for parametric sysid (ablation) | parametric sysid · parametric environments | cost | No clear effect |
| use 5 implicit planning steps per iteration (ablation) | end-to-end algorithm · end-to-end | cost | Improved |
Closest papers by meaning
- DiffTORI: Differentiable Trajectory Optimization for Deep Reinforcement and Imitation Learning (NeurIPS 2024 · similarity 0.87)
- Langevin Rollout Optimization for Modelic Reinforcement Learning (ICML 2026 · similarity 0.86)
- Learning Self-Imitating Diverse Policies (ICLR 2019 · similarity 0.86)
- PWM: Policy Learning with Multi-Task World Models (ICLR 2025 · similarity 0.85)
- Trajectory World Models for Heterogeneous Environments (ICML 2025 · similarity 0.85)
- Gradient Descent as Loss Landscape Navigation: a Normative Framework for Deriving Learning Rules (NeurIPS 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.
