Reinforcement Learning Based Multi-Terrain Adaptive Whole-Body Control for a Quadruped Manipulator

Yanzhou Jin; Fei Suo; Chang Liu; Yasuhisa Hasegawa; Liming Chen; Yusuke Iwasawa; Yutaka Matsuo; Yaonan Zhu

cs.ROOct 16, 2026

Reinforcement Learning Based Multi-Terrain Adaptive Whole-Body Control for a Quadruped Manipulator

Authors: Yanzhou Jin, Fei Suo, Chang Liu, Yasuhisa Hasegawa, Liming Chen, Yusuke Iwasawa, Yutaka Matsuo, Yaonan Zhu

Organizations: Graduate School of Engineering, The University of Tokyo, Japan. · Department of Micro-Nano Mechanical Science and Engineering, Nagoya University, Nagoya, 464-8603, Japan. · Department of Mathematics and Informatics, Ecole Centrale de Lyon, France.

Abstract

Legged robots equipped with manipulators offer promising capabilities for loco-manipulation tasks in unstructured environments. However, achieving task-consistent end-effector behavior during locomotion remains challenging due to terrain-induced disturbances and the strong coupling between locomotion and manipulation, particularly for small-scale quadruped platforms. A reinforcement-learning based whole-body control framework is proposed for quadruped loco-manipulation on complex terrain. The proposed framework combines a hybrid task representation with task-prioritized whole-body learning to coordinate locomotion and manipulation while preserving task-consistent end-effector behavior. A lightweight quadruped manipulator platform is further developed to validate the proposed framework on a small-scale quadruped system. Real-world experiments demonstrate successful deployment on complex terrains such as slopes up to 30 degrees, stairs, and deformable terrain. These results demonstrate the feasibility of deploying learned whole-body control policy for quadruped loco-manipulation in real-world environments and highlight the importance of task-consistent whole-body coordination for terrain-adaptive loco-manipulation.

Figures & tables

Fig. 1: Manipulator-induced mass shifts and complex terrain interaction jointly challenge rough-terrain loco-manipulation on small-scale quadruped robots. The proposed whole-body control framework enables coordinated terrain traversal and manipulation while preserving task-consistent end-effector behavior.

Fig. 2: Overview of the proposed framework. (a) Customized lightweight 6-DoF manipulator and the quadruped manipulator system for challenging terrain loco-manipulation. (b)-(d) Comparison of body-frame, world-frame, and hybrid task representations for rough-terrain loco-manipulation. (e) System architecture comprising semantic perception, hybrid task representation, prioritized whole-body learning, and real-world deployment.

Fig. 3: Ablation and performance evaluation of the proposed framework. Left: (a)-(h) Representative behaviors of the full policy across terrains at difficulty 1.0. (i)-(l) Distribution of locomotion and manipulation errors for different ablated variants (No P1-P3: corresponding Priority I-III regulation components disabled.). (m) locomotion-manipulation trade-off for the full policy and its ablations. Right: Base Motion Error, EE Tracking Error, and joint overload exposure across terrain difficulty for WBC and the baselines. Lower values indicate better performance.

Fig. 4: Representative rough-terrain traversal demonstrations. Blue lines indicate the horizontal direction of the end-effector. The proposed whole-body controller maintains a nearly level end-effector orientation during stair and slope traversal under different arm configurations, whereas the built-in MPC baseline exhibits larger orientation deviations and degraded terrain traversal performance.

Fig. 5: End-effector tracking on stairs and under manual disturbances. EE tracking errors are estimated via FK using measured joint states, with body and foot-plane tilts estimated from projected gravity and FK-based foot positions, respectively.

Fig. 6: Autonomous rough-terrain grasping task demonstrations. Left: execution sequences for the four target objects, each showing the robot traversing the terrain, approaching the table, and lifting the object; the top row is the deformable terrain, the bottom row the lateral stair descent, where the four feet occupy footholds at different heights. Right: Sankey diagram of the eight failed trials, grouped by failure stage.

Ref	Terrain Demonstration	Task-space Formulation	System Scale (kg)	Manipulation on Rough Terrain
RAKOMO [ 15 ]	Flat Ground	World-oriented	$\sim$ 90	$\times$
RoLoMa [ 4 ]	Ramp	World-oriented	$>$ 50	$\times$
DeepWBC [ 23 ]	Rough Terrain	Body-oriented	$\sim$ 14.8	$\times$
Visual WBC [ 21 ]	Rough Terrain	Body-oriented	$\sim$ 60	$\times$
MLM [ 22 ]	Stairs / Obstacles	World-oriented	19.9	$\checkmark$
UMI-on-Legs [ 8 ]	Outdoor Terrain	World-oriented	$\sim$ 18.4	$\triangle$

TABLE I: Qualitative comparison with representative quadruped loco-manipulation systems.

Legged Robotics

Legged robotics studies how robots with legs move, perceive their surroundings, and perform useful tasks across varied terrain.

See all 118 papers

Learning Terrain-Aware Whole-Body Control for Perceptive Legged Loco-Manipulation
May 29, 2026Sikai Guo, Yudong Zhong, Guoyang Zhao +3Legged Robotics Terrain-Aware Robot Locomotion
Robotics and Autonomous Systems Thrust, The Hong Kong University of Science and Technology (Guangzhou), Guangzhou 511453, China
Learning Holistic Whole-Body Loco-Manipulation with a Bipedal Mobile Manipulator
Sep 16, 2026Zhongyu Chen, Yuxuan Nai, Qian Chen +8RL for Robotics Humanoid Loco-Manipulation
ZJU-UIUC Institute. · LimX Dynamics.
Sumo: Dynamic and Generalizable Whole-Body Loco-Manipulation
Apr 9, 2026John Z. Zhang, Maks Sorokin, Jan Brüdigam +14Generalization in Robotic Manipulation Whole-Body Robotic Manipulation
MIT · RAI Institute · Cornell University

May 29, 2026J/K move · Enter open · S save

Learning Terrain-Aware Whole-Body Control for Perceptive Legged Loco-Manipulation

Sikai Guo, Yudong Zhong, Guoyang Zhao, Botao Dang+2

Robotics and Autonomous Systems Thrust, The Hong Kong University of Science and Technology (Guangzhou), Guangzhou 511453, China

Legged Robotics Terrain-Aware Robot Locomotion RL for Legged Locomotion

Open paper AskPDF arXiv

Legged manipulators integrate exceptional terrain adaptability along with mobile manipulation capabilities, which make them highly promising for deployment in human-centric environments. By coordinating the control of both legs and arms, a whole-body controller can significantly expand the operational workspace of legged manipulators. However, many existing whole-body controllers primarily depend on proprioception and do not incorporate the critical exteroception required for effective terrain topology perception. This limitation can hinder their ability to adapt to varying environmental conditions and navigate complex terrains effectively. In this paper, we introduce TA-WBC, a terrain-aware whole-body control framework for legged manipulators, which features a novel RL-based unified policy tailored to whole-body loco-manipulation tasks in various terrains. Specifically, we employ a \rev{hierarchical exteroceptive encoder} to extract terrain features, providing an essential basis for the robot to proactively adapt posture and footholds. Furthermore, to facilitate stable cross-terrain loco-manipulation, we propose a novel end-effector sampling method based on the foot contact plane, \rev{decoupling the manipulation target from base height, roll, and pitch variations}. Moreover, a dual-policy distillation module is introduced to integrate expansive whole-body motion with terrain adaptability without catastrophic forgetting. The simulation and real-world experiments validate the robustness of our proposed controller, which leads to a larger reachable space, less tracking error, and reduced unexpected stumbles. This unified policy highlights the promising capabilities of legged manipulators in performing loco-manipulation tasks across complex terrains.

Abstract

Legged Robotics

Explore similar work

Learning Terrain-Aware Whole-Body Control for Perceptive Legged Loco-Manipulation

Learning Holistic Whole-Body Loco-Manipulation with a Bipedal Mobile Manipulator

Sumo: Dynamic and Generalizable Whole-Body Loco-Manipulation