aiqbench analysis
The paper proposes FOCUS, an imitation-learning method that learns from state-only demonstrations by aligning short rollout windows to sampled expert segments with optimal transport, then training a differentiable actor-critic with a learned critic. The reported gains are in sample efficiency and training stability on contact-rich locomotion tasks in DFlex, including a 152-muscle humanoid, compared with adversarial and prior differentiable baselines. These are simulation results, so readers should check how well they transfer to real robots and to settings where expert data is noisier or less complete.
FOCUS: First-Order Control Using Optimal Transport Supervision for Imitation from Observations
arXiv:2610.41540v1 Announce Type: new Abstract: Imitation learning from observations must recover behavior from state-only demonstrations without known temporal correspondence. Differentiable simulators provide pathwise policy gradients, but differentiating through long, contact-rich trajectories can make optimization unstable. We introduce FOCUS, which combines local optimal-transport alignment with short-horizon differentiable actor-critic learning. FOCUS matches agent rollout windows to multiple sampled expert segments using a soft best-of-$K$ objective that emphasizes locally compatible matches. The resulting transport-based costs provide differentiable imitation rewards, while a learned critic bootstraps value beyond the rollout horizon. FOCUS requires neither temporally aligned trajectories nor expert-aligned initialization or teacher forcing. Across contact-rich locomotion tasks in DFlex, including a musculoskeletal humanoid controlled by 152 muscle activations, FOCUS improves sample efficiency and training stability over adversarial imitation-from-observation methods and prior differentiable imitation baselines.
Abstract:Imitation learning from observations must recover behavior from state-only demonstrations without known temporal correspondence. Differentiable simulators provide pathwise policy gradients, but differentiating through long, contact-rich trajectories can make optimization unstable. We introduce FOCUS, which combines local optimal-transport alignment with short-horizon differentiable actor-critic learning. FOCUS matches agent rollout windows to multiple sampled expert segments using a soft best-of-$K$ objective that emphasizes locally compatible matches. The resulting transport-based costs provide differentiable imitation rewards, while a learned critic bootstraps value beyond the rollout horizon. FOCUS requires neither temporally aligned trajectories nor expert-aligned initialization or teacher forcing. Across contact-rich locomotion tasks in DFlex, including a musculoskeletal humanoid controlled by 152 muscle activations, FOCUS improves sample efficiency and training stability over adversarial imitation-from-observation methods and prior differentiable imitation baselines.
Machine Learning (cs.LG); Robotics (cs.RO)
arXiv:2610.41540 [cs.LG]
(or arXiv:2610.41540v1 [cs.LG] for this version)
https://doi.org/10.48550/arXiv.2610.41540 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 220 words
Related coverage
What You See Is What You Get: Observation-Aligned Supervision for Chart-to-Code Generation
arXiv · Jul 28, 2026
Recovering Physical Dynamics from Discrete Observations via Intrinsic Differential Consistency
arXiv · Sep 30, 2026
Acting Through Registers
arXiv · Oct 14, 2026
EgoMoD: Predicting Global Maps of Dynamics from Local Egocentric Observations
arXiv · Jun 12, 2026