mlsanityData to 14 Oct 2026

A Population-Level Analysis of Neural Dynamics in Robust Legged Robots

arXiv 2306.15793 ↗First posted 27 Jun 2023cs.RO cs.AI cs.LGcited by 0

Not on the map. Why?

Abstract (arXiv)

Recurrent neural network-based reinforcement learning systems are capable of complex motor control tasks such as locomotion and manipulation, however, much of their underlying mechanisms still remain difficult to interpret. Our aim is to leverage computational neuroscience methodologies to understanding the population-level activity of robust robot locomotion controllers. Our investigation begins by analyzing topological structure, discovering that fragile controllers have a higher number of fixed points with unstable directions, resulting in poorer balance when instructed to stand in place. Next, we analyze the forced response of the system by applying targeted neural perturbations along directions of dominant population-level activity. We find evidence that recurrent state dynamics are structured and low-dimensional during walking, which aligns with primate studies. Additionally, when recurrent states are perturbed to zero, fragile agents continue to walk, which is indicative of a stronger reliance on sensory input and weaker recurrence.

Key facts. No citing papers in the map yet (data through 14 Oct 2026). Builds on 30 of its 42 references that are in the map. First posted on arXiv 27 Jun 2023.

Builds on

30 of 42 references.

References · 30

  1. 2014Adam: A Method for Stochastic Optimizationcited by 37,240
  2. 2017Proximal Policy Optimization Algorithmscited by 15,215
  3. 2021Isaac Gym: High Performance GPU-Based Physics Simulation For Robot Learningcited by 998
  4. 2018QT-Opt: Scalable Deep Reinforcement Learning for Vision-Based Robotic Manipulationcited by 743
  5. 2019Solving Rubik's Cube with a Robot Handcited by 734
  6. 2018Learning Dexterous In-Hand Manipulationcited by 687
  7. 2020Learning Quadrupedal Locomotion over Challenging Terraincited by 604
  8. 2021Learning to Walk in Minutes Using Massively Parallel Deep Reinforcement Learningcited by 512
  9. 2022Learning robust perceptive locomotion for quadrupedal robots in the wildcited by 382
  10. 2016Deep Reinforcement Learning for Robotic Manipulation with Asynchronous Off-Policy Updatescited by 353
  11. 2017Asymmetric Actor Critic for Image-Based Robot Learningcited by 238
  12. 2022DeXtreme: Transfer of Agile In-hand Manipulation from Simulation to Realitycited by 149
18 more
  1. 2018Learning Synergies between Pushing and Grasping with Self-supervised Deep Reinforcement Learningcited by 128
  2. 2022Deep Whole-Body Control: Learning a Unified Policy for Manipulation and Locomotioncited by 121
  3. 2020Sim-to-Real Learning of All Common Bipedal Gaits via Periodic Reward Compositioncited by 96
  4. 2019Universality and individuality in neural dynamics across large populations of recurrent networkscited by 61
  5. 2018Emergence of grid-like representations by training recurrent neural networks to perform spatial localizationcited by 52
  6. 2022Advanced Skills by Learning Locomotion and Local Navigation End-to-Endcited by 52
  7. 2022GenLoco: Generalized Locomotion Controllers for Quadrupedal Robotscited by 48
  8. 2023Barkour: Benchmarking Animal-level Agility with Quadruped Robotscited by 41
  9. 2019Reverse engineering recurrent networks for sentiment classification reveals line attractor dynamicscited by 39
  10. 2022Advanced Skills through Multiple Adversarial Motion Priors in Reinforcement Learningcited by 37
  11. 2020Learning Memory-Based Control for Human-Scale Bipedal Locomotioncited by 28
  12. 2019Deep neuroethology of a virtual rodentcited by 23
  13. 2020How recurrent networks implement contextual processing in sentiment analysiscited by 11
  14. 2019Emergence of functional and structural properties of the head direction system by optimization of recurrent neural networkscited by 9
  15. 2020The geometry of integration in text classification RNNscited by 9
  16. 2021Recurrent neural network models for working memory of continuous variables: activity manifolds, connectivity patterns, and dynamic codescited by 8
  17. 2021Emergent behavior and neural dynamics in artificial agents tracking turbulent plumescited by 2
  18. 2023From Data-Fitting to Discovery: Interpreting the Neural Dynamics of Motor Control through Reinforcement Learningcited by 1

Built on by

Not cited yet.