Research map / Reinforcement learning / Visual and representation RL
Accelerated co-design of robots through morphological pretraining
Luke Strgar, Sam Kriegman
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 |
|---|---|---|---|
| generational finetuning of the pretrained controller during evolution (few-shot evolution) (ablation) | zero-shot evolution · Phototaxis evolution, population diversity | mean pairwise Hamming distance diversity | Improved |
| add morphological pretraining before evolution instead of co-designing from scratch (ablation) | simultaneous co-design · Phototaxis evolution, diversity | population diversity | Improved |
| freeze pretrained controller (zero-shot evolution, no finetuning) (ablation) | few-shot evolution · Phototaxis evolution, diversity | population diversity | Hurt |
| pretrain universal controller with differentiable simulation gradients | MLP universal controller · Phototaxis, training distribution | final/initial distance loss: 1.0 -> 0.3 | Improved |
| use pretrained controller for evolution without co-training (ablation) | design evolution · Phototaxis evolution, wall-clock | time to converge: 17 minutes vs 109 minutes | Improved |
| crossover between designs under simultaneous co-design | simultaneous co-design from scratch · Phototaxis evolution | design quality vs random designs | No clear effect |
| crossover with a pretrained universal controller | zero-shot evolution · Phototaxis evolution, offspring survival | offspring survival | Improved |
Closest papers by meaning
- Convergent Functions, Divergent Forms (NeurIPS 2025 · similarity 0.91)
- BodyGen: Advancing Towards Efficient Embodiment Co-Design (ICLR 2025 · similarity 0.90)
- Embracing Evolution: A Call for Body-Control Co-Design in Embodied Humanoid Robot (NeurIPS 2026 · similarity 0.89)
- DittoGym: Learning to Control Soft Shape-Shifting Robots (ICLR 2024 · similarity 0.89)
- Distilling Morphology-Conditioned Hypernetworks for Efficient Universal Morphology Control (ICML 2024 · similarity 0.88)
- MorphVAE: Advancing Morphological Design of Voxel-Based Soft Robots with Variational Autoencoders (AAAI 2024 · similarity 0.87)
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.
