aiqbench analysis
MIMIC-MJX represents a significant step in inverse reinforcement learning by using neural networks to infer motor control policies directly from kinematic data, bridging the gap between observed behavior and underlying biological mechanisms. This framework enables high-fidelity simulation of animal locomotion across diverse biomechanical models with minimal training data, offering new tools for neuroscience research and robotic biomimicry.
MIMIC-MJX: Neuromechanical Emulation of Animal Behavior
arXiv:2511.20532v3 Announce Type: replace-cross Abstract: The primary output of the nervous system is movement and behavior. While recent advances have democratized pose tracking during complex behavior, kinematic trajectories alone provide only indirect access to the underlying control processes. Here we present MIMIC-MJX, a framework for learning biomechanically grounded neural control policies from kinematics. MIMIC-MJX provides a platform for modeling the generative process of motor control by training neural controllers that learn to actuate biomechanical animal models in physics simulation to reproduce real kinematic trajectories. We demonstrate that our implementation is accurate, fast, and generalizable to diverse animal body models, and that it can be trained with modest amounts of motion data. MIMIC-MJX can be used to model motor control policies and simulate behavioral experiments, illustrating its potential as an integrative modeling framework for neuroscience.
Authors:Charles Y. Zhang (1), Yuanjia Yang (2, 3), Aidan Sirbu (4, 5), Elliott T.T. Abe (6), Emil Wärnberg (1), Eric J. Leonardis (2), Diego E. Aldarondo (1), Adam Lee (1, 2), Aaditya Prasad (7), Jason Foat (2), Kaiwen Bian (2), Joshua Park (2), Rusham Bhatt (2), Vyom N. Patel (8), Hutton Saunders (2), Austin O. Barbano (2), Akira Nagamori (2), Ayesha R. Thanawalla (2), Kee Wui Huang (2), Fabian Plum (9), Hendrik K. Beck (9), Steven W. Flavell (7, 10), David Labonte (9), Blake A. Richards (4, 5, 11), Bingni W. Brunton (6), Eiman Azim (2), Bence P. Ölveczky (1), Talmo D. Pereira (2) ((1) Harvard University, (2) Salk Institute for Biological Studies, (3) University of California San Diego, (4) Mila, (5) McGill University, (6) University of Washington, (7) Massachusetts Institute of Technology, (8) Neuromatch, (9) Imperial College London, (10) Howard Hughes Medical Institute, (11) Canadian Institute for Advanced Research)
Abstract:The primary output of the nervous system is movement and behavior. While recent advances have democratized pose tracking during complex behavior, kinematic trajectories alone provide only indirect access to the underlying control processes. Here we present MIMIC-MJX, a framework for learning biomechanically grounded neural control policies from kinematics. MIMIC-MJX provides a platform for modeling the generative process of motor control by training neural controllers that learn to actuate biomechanical animal models in physics simulation to reproduce real kinematic trajectories. We demonstrate that our implementation is accurate, fast, and generalizable to diverse animal body models, and that it can be trained with modest amounts of motion data. MIMIC-MJX can be used to model motor control policies and simulate behavioral experiments, illustrating its potential as an integrative modeling framework for neuroscience.
Neurons and Cognition (q-bio.NC); Artificial Intelligence (cs.AI); Robotics (cs.RO)
arXiv:2511.20532 [q-bio.NC]
(or arXiv:2511.20532v3 [q-bio.NC] for this version)
Full text under the publisher's open licence · 296 words
Related coverage
The J-Space Debate, Agent Swarms, and Pacing Frontier AI - Digital Minds Newsletter #4
LessWrong · Sep 18, 2026
The embodied brain: Bridging the brain, body, and behavior with biorealistic neuromechanical models
arXiv · Jul 21, 2026
The embodied brain: Bridging the brain, body, and behavior with biorealistic neuromechanical models
arXiv · Jun 16, 2026
Tech to Replace Animal Testing Is Almost Ready. Scientists Are Not
IEEE Spectrum · Sep 29, 2026