aiqbench analysis
This paper applies reinforcement learning to quadruped robot control, and its main contribution is a reward design that separates task goals, operational constraints, energy use, and terrain perception. The reported results are a 56% reduction in cost of transport and a 96% reduction in operational-limit violations versus a complex-reward baseline, with zero-shot transfer to a physical Unitree Go2. Readers should note these gains are measured against one baseline, and the transfer claim rests on a single robot platform.
LoComposition: Terrain-Adaptive Energy-Efficient Quadruped Locomotion without Gait Priors
arXiv:2606.15896v2 Announce Type: replace-cross Abstract: Learning-based quadrupedal locomotion typically relies on complex reward formulations that entangle task specification, operational limits, gait preference, and terrain adaptation within a single optimization objective. We instead treat these functions through distinct mechanisms: rewards for task specification, constraints for operational limits, energy minimization for gait preference, and exteroceptive perception for adapting energy use to terrain difficulty. We show that these components jointly enable efficient, terrain-adaptive locomotion, and that removing each component exposes a distinct failure mode. Our formulation removes explicit gait priors (including air-time, contact-count, and foot-clearance targets) in favor of emergent behavior. Compared to a conventional complex-reward baseline, our formulation achieves comparable terrain traversal while reducing cost of transport by 56% and operational-limit violations by 96%. The resulting policies transfer zero-shot to a physical Unitree Go2 using LiDAR-based elevation mapping. Project website with videos: https://tinyurl.com/locomposition.
Abstract:Learning-based quadrupedal locomotion typically relies on complex reward formulations that entangle task specification, operational limits, gait preference, and terrain adaptation within a single optimization objective. We instead treat these functions through distinct mechanisms: rewards for task specification, constraints for operational limits, energy minimization for gait preference, and exteroceptive perception for adapting energy use to terrain difficulty. We show that these components jointly enable efficient, terrain-adaptive locomotion, and that removing each component exposes a distinct failure mode. Our formulation removes explicit gait priors (including air-time, contact-count, and foot-clearance targets) in favor of emergent behavior. Compared to a conventional complex-reward baseline, our formulation achieves comparable terrain traversal while reducing cost of transport by 56% and operational-limit violations by 96%. The resulting policies transfer zero-shot to a physical Unitree Go2 using LiDAR-based elevation mapping. Project website with videos: this https URL.
Robotics (cs.RO); Machine Learning (cs.LG)
arXiv:2606.15896 [cs.RO]
(or arXiv:2606.15896v2 [cs.RO] for this version)
https://doi.org/10.48550/arXiv.2606.15896 Focus to learn more arXiv-issued DOI via DataCite
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 · 232 words
Related coverage
LoComposition: Terrain-Adaptive Energy-Efficient Quadruped Locomotion without Gait Priors
arXiv · Jun 16, 2026
Experience-Driven Continual Learning of Terrain Traversability for Quadruped Robots
arXiv · Oct 1, 2026
Energy-Efficient Gait Adaptation via Hierarchical Reinforcement Learning for Quadrupedal Locomotion Across Diverse Terrains
arXiv · Oct 8, 2026
Neuromorphic Reinforcement Learning for Quadruped Locomotion Control on Uneven Terrain
arXiv · Jun 10, 2026