Research map / Reinforcement learning / Visual and representation RL
Exo-Plore: Exploring Exoskeleton Control Space through Human-aligned Simulation
Geonho Leem, Jaedong Lee, Jehee Lee, Seungmoon Song, Jungdam Won
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 the human-exoskeleton interaction (HEI) reward based on resistance minimization (ablation) | Exo-Plore gait data generator · 4 km/h assistive power vs delay | Pearson correlation with human data: 0.69 -> 0.83 | Improved |
| use HEI reward instead of assistance maximization r_assist (ablation) | Exo-Plore gait data generator · 4 km/h assistive power vs delay | Pearson correlation with human data: 0.23 -> 0.83 | Improved |
| add HEI reward (versus none or assistance maximization) (ablation) | Exo-Plore · maximum metabolic reduction rate | alignment to human data | Improved |
| add intramuscular regularization (IMR) to the muscle coordination network (ablation) | MCN · muscle activation training | physiological plausibility of segment activations | Improved |
| use neural network surrogate instead of Gaussian process (ablation) | exoskeleton optimizer surrogate · peak memory, 10,000 samples | peak memory (MB): 2232 -> 360 | Improved |
| increase surrogate training set from 5k/10k to 20k+ samples (ablation) | NN surrogate · CoT landscape | smoothness and plausibility of optimized parameters: 5k, 10k noisy -> 20k, 60k smoother | Improved |
| use Huber loss and Latin Hypercube Sampling instead of MSE and uniform sampling (ablation) | NN surrogate · CoT landscape (N=60k) | landscape smoothness | Improved |
| increase gradient penalty from lambda_gp 0.01 to 0.1 (ablation) | NN surrogate · CoT landscape | optimization behavior: lambda_gp 0.01 vs 0.1 | Mixed |
| choose MEE exponent alpha = 1.5 (ablation) | metabolic energy expenditure · preferred walking speed | PWS vs human 1.25 m/s: alpha = 1.5 closest to v_real = 1.25 m/s | Improved |
Closest papers by meaning
- ReMAP: Neural Model Reprogramming with Network Inversion and Retrieval-Augmented Mapping for Adaptive Motion Forecasting (NeurIPS 2024 · similarity 0.82)
- Reinforcement Learning Control of a Physical Robot Device for Assisted Human Walking without a Simulator (ICML 2025 · similarity 0.82)
- Generating Freeform Endoskeletal Robots (ICLR 2025 · similarity 0.81)
- CrossLoco: Human Motion Driven Control of Legged Robots via Guided Unsupervised Reinforcement Learning (ICLR 2024 · similarity 0.81)
- SUGAR: A Scalable Human-Video-Driven Generalizable Humanoid Loco-Manipulation Learning Framework (NeurIPS 2026 · similarity 0.81)
- Adversarial Locomotion and Motion Imitation for Humanoid Policy Learning (NeurIPS 2025 · similarity 0.80)
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.
