grad Mimic: Ultra-Efficient Humanoid Control with Differentiable Simulation
cs.RO, cs.LG
Submitted: 2026-10-30
Updated: 2026-10-30
Comments: 26 pages, 7 figures
Code: https://github.com/OamIcs/NablaMimic
Project page: https://oamics.github.io/NablaMimic
License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
Terminology
Sources
- Deep Learning with Differential Privacy
- Maximum a Posteriori Policy Optimisation
- Differentially Private Learning with Adaptive Clipping
- GMT: General Motion Tracking for Humanoid Whole-Body Control
- Model-Augmented Actor-Critic: Backpropagating through Paths
- A Differentiable Physics Engine for Deep Learning in Robotics
- PWM: Policy Learning with Multi-Task World Models
- Dream to Control: Learning Behaviors by Latent Imagination
- Learning Continuous Control Policies by Stochastic Value Gradients
- Adam: A Method for Stochastic Optimization
- Revisiting Gradient Clipping: Stochastic bias and tight convergence guarantees
- Target-based Surrogates for Stochastic Optimization
- SuperMimic: Scaling Humanoid Whole-Body Control with One Differentiable World Model
- Do Transformer World Models Give Better Policy Gradients?
- Robust Stochastic Optimization via Gradient Quantile Clipping
- Isaac Lab: A GPU-Accelerated Simulation Framework for Multi-Modal Robot Learning
- On the difficulty of training Recurrent Neural Networks
- Advantage-Weighted Regression: Simple and Scalable Off-Policy Reinforcement Learning
- Proximal Policy Optimization Algorithms
- A Tail-Index Analysis of Stochastic Gradient Noise in Deep Neural Networks
Related papers
- Rethinking Actions: Continuity-Aware Priors for World-Model Planning
- MOSAIC: Modular Obstacle and Spatial-Aware Framework for Manipulation In Cluttered Scenes
- SAGAS: Semantic-Aware Graph-Assisted Stitching for Offline Temporal Logic Planning
- DexSeed: Expanding Human Hand-Object Interactions into Robot Demonstrations for Dexterous Policy Learning
- LaST0.5: A Generalist World Action Model with Interaction-Centric Latent Reasoning
- CIALLO: Learning Shared Intent Representations for Cross-Embodiment Manipulation