Research map / Reinforcement learning / Policy gradient and actor-critic
Efficient Morphology–Control Co-Design via Stackelberg PPO under Non-Differentiable Leader–Follower Interfaces
Yanning Dai, Yuhui Wang, Dylan R. Ashley, Jürgen Schmidhuber
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 |
|---|---|---|---|
| add Stackelberg implicit differentiation to the leader gradient (vs BodyGen) | BodyGen · Stepper-Regular | final episode reward: 4685.94 -> 7215.20 | Improved |
| add Stackelberg gradient on challenging 3D tasks | BodyGen · Crawler, Stepper-Regular, Stepper-Hard, Pusher | final episode reward: +32.02% | Improved |
| use Stackelberg PPO over strongest baseline across all environments | BodyGen · all nine environments | final episode reward: +20.66% | Improved |
| add Stackelberg gradient on Swimmer | BodyGen · Swimmer | final episode reward: 1302.64 -> 1334.98 | No clear effect |
| add PPO clipping to SID (full vs SID-only) (ablation) | SID estimator · Stepper-Regular | final episode reward: 1257.33 -> 7215.20 | Improved |
| add SID to PPO (full vs PPO-only) (ablation) | PPO · Stepper-Regular | final episode reward: 4685.94 -> 7215.20 | Improved |
| use Fisher approximation instead of analytic Hessian (ablation) | Stackelberg PPO · Stepper-Regular | final episode reward: 2500 -> 6000 | Improved |
| use intermediate lambda (0.5 to 10) instead of lambda 0 or infinity (ablation) | Stackelberg PPO · Stepper-Regular | final episode reward | Improved |
| add PPO clipping at epsilon 0.2 instead of no clipping (ablation) | Stackelberg PPO · Stepper-Regular | final episode reward: 1233.26 -> 7215.20 | Improved |
| raise clipping threshold from 0.2 to 0.4 (ablation) | Stackelberg PPO · Stepper-Regular | final episode reward: 7215.20 -> 7907.01 | Improved |
| raise clipping threshold from 0.4 to 0.8 (ablation) | Stackelberg PPO · Stepper-Regular | final episode reward: 7907.01 -> 2656.92 | Hurt |
| raise leader horizon T from 3 to 5 (ablation) | Stackelberg PPO · Stepper-Regular | final episode reward: 6188.99 -> 7215.20 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- BodyGen — BodyGen: Advancing Towards Efficient Embodiment Co-Design (ICLR 2025) · better on 9 of 9 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Convergent Functions, Divergent Forms (NeurIPS 2025 · similarity 0.88)
- BodyGen: Advancing Towards Efficient Embodiment Co-Design (ICLR 2025 · similarity 0.87)
- Debate2Create: Robot Co-design via Multi-Agent LLM Debate (ICML 2026 · similarity 0.86)
- Embracing Evolution: A Call for Body-Control Co-Design in Embodied Humanoid Robot (NeurIPS 2026 · similarity 0.86)
- DittoGym: Learning to Control Soft Shape-Shifting Robots (ICLR 2024 · similarity 0.83)
- CaPo: Cooperative Plan Optimization for Efficient Embodied Multi-Agent Cooperation (ICLR 2025 · similarity 0.83)
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.
