Stay Seated: Learning Omnidirectional Humanoid Locomotion on a Passive Mobile Chair with Casters

Kango Yanagida; Kazuki Miyazawa; Takato Horii

cs.ROAug 28, 2026

Stay Seated: Learning Omnidirectional Humanoid Locomotion on a Passive Mobile Chair with Casters

Authors: Kango Yanagida, Kazuki Miyazawa, Takato Horii

Organizations: Dept. of Systems Innovation, Graduate School of Engineering Science, The University of Osaka, Japan. · International Research Center for Neurointelligence, The University of Tokyo, Japan.

Abstract

Humanoid robots with quasi-direct-drive actuators continuously generate joint torque while standing, whereas seated humans delegate weight support to chairs during desk work. As a first step toward seated loco-manipulation, we study omnidirectional seated locomotion on a passive mobile chair, requiring unfixed pelvis-seat contact and intermittent foot-floor propulsion of the robot-chair system. We extend a standard standing velocity-tracking environment with a passive-chair model, seated-state rewards, critic-only chair observations, and task-tailored contact settings. The policy is learned without motion-imitation rewards; its actor uses only proprioception and velocity commands, without contact sensing or chair states. In random-command evaluation, the policies tracked omnidirectional commands through nearly all 20-s rollouts, and the best seated policies could outperform the Standing policy in velocity tracking. Across four training seeds, a $2^3$ full-factorial comparison of symmetry regularization (SY), foot-slip regularization (FS), and command curriculum (CC) showed that FS reduced CoT but increased tracking error and that some FS-only policies converged to stationary local optima. Combining FS with either SY or CC avoided this failure without retuning FS, while SY improved bilateral leg symmetry during longitudinal motion. Direction-resolved analysis showed CoT ordered backward $<$ lateral $\ll$ forward, with planted-leg extension in backward and lateral motion and knee flexion following heel contact in forward motion. The learned policy achieved zero-shot sim-to-real transfer to a Unitree G1 and generated omnidirectional seated locomotion.

Figures & tables

Fig. 1 : Seated locomotion on a passive mobile chair with casters during translation and turning.

Fig. 2 : Overview of the seated locomotion learning framework, including the asymmetric actor–critic architecture, refined pelvis collision mesh, and passive-chair model. The full-factorial study varies command curriculum (CC), symmetry regularization (SY), and foot-slip penalty (FS).

Type	Component	Space
Actor and critic observations	Torso angular velocity	$\mathbb{R}^{3}$
	Projected gravity	$\mathbb{R}^{3}$
	Joint positions / velocities	$\mathbb{R}^{29}/\mathbb{R}^{29}$
	Last action	$\mathbb{R}^{29}$
	Velocity command	$\mathbb{R}^{3}$
Critic-only observation	Torso linear velocity	$\mathbb{R}^{3}$

TABLE I : Observation and action spaces for the seated locomotion task.

Table 4

Condition	Tracking RMSE $v_{x}$ [m/s]	Tracking RMSE $v_{y}$ [m/s]	Tracking RMSE $\omega_{z}$ [rad/s]	Timeout success [%]	Seat-relative RMS [m]	Torso-tilt RMS [rad]
Baseline	$0.1650\pm 0.0084$	$0.1441\pm 0.0071$	$0.1506\pm 0.0014$	$99.45\pm 0.17$	$0.0153\pm 0.0008$	$0.0459\pm 0.0018$
SY	$0.1523\pm 0.0043$	$\underline{0.1284\pm 0.0033}$	$0.1444\pm 0.0006$	$99.68\pm 0.52$	$0.0157\pm 0.0008$	$\underline{0.0450\pm 0.0019}$
FS	$0.1964\pm 0.0432$	$0.1879\pm 0.0473$	$\mathbf{0.1411\pm 0.0099}$	$99.53\pm 0.17$	$\mathbf{0.0151\pm 0.0009}$	$0.0480\pm 0.0038$
CC	$0.1524\pm 0.0023$	$0.1341\pm 0.0024$	$0.1498\pm 0.0021$	$99.73\pm 0.17$	$\underline{0.0152\pm 0.0011}$	$0.0456\pm 0.0025$
SY+FS	$0.1542\pm 0.0028$	$0.1382\pm 0.0033$	$0.1457\pm 0.0028$	$99.68\pm 0.17$	$0.0159\pm 0.0012$	$0.0469\pm 0.0035$
SY+CC	$\mathbf{0.1512\pm 0.0002}$	$\mathbf{0.1268\pm 0.0036}$	$0.1453\pm 0.0025$	$\mathbf{99.80\pm 0.14}$	$0.0156\pm 0.0023$	$\mathbf{0.0436\pm 0.0030}$

TABLE IV : Random-command evaluation. Values are the mean and standard deviation across four training seeds; each learned policy was evaluated over 1,000 rollouts. In each column, bold and underlined values denote the best and second-best values among the eight training conditions, respectively.

Fig. 3 : Translational tracking error heatmaps from the fixed-command evaluation for FS, SY+FS, FS+CC and SY+FS+CC.

Fig. 4 : Command area as a function of the tracking-error threshold. Curves and shaded regions show the mean and standard deviation across four training seeds, respectively.

Fig. 5 : Example simulation rollout during a velocity-command transition from rightward to forward motion. Purple and cyan arrows indicate the commanded and measured planar velocities, respectively. The policy remained seated and adjusted its motion direction after the command switch.

TABLE V : Foot contact characteristics across movement directions. Contact fraction denotes left-only/right-only contact. Contact force denotes the resultant floor-contact force averaged during contact for the left/right foot.

TABLE VI : Direction- and speed-dependent tracking, energy efficiency, and contact requirements. Contact fraction reports the mean single-foot/flight fraction, and contact force reports tangential/normal force on the floor.

Panel of Figure 11 — Fig. 6 : Snapshots from sim-to-real deployment of the seated locomotion policy on a physical Unitree G1.

Fig. 7 : Battery-side power; dotted lines show means.

Humanoid Robot Locomotion

Humanoid robot locomotion studies how bipedal robots walk, run, recover from disturbances, and traverse varied terrain.

See all 146 papers

HuMBLE: Human Motion-Driven Behavior Learning for Embodied Locomotion
Oct 7, 2026Mike Zhang, Dongho Kang, Kevin Bergamin +19Human-to-Robot Motion Imitation Humanoid Robot Locomotion
RAI Institute, USA · Boston Dynamics, USA · Carnegie Mellon University, USA
Natural Sit-to-Stand Motion Synthesis For Humanoids via Guided Assistance Curricula and Staged Rewards
Aug 21, 2026Meet Pal Singh, Vyankatesh Ashtekar, Ashish DuttaHumanoid Robot Control Humanoid Robotics
Department of Mechanical Engineering, Indian Institute of Technology Kanpur, Kanpur, Uttar Pradesh, India
Partial Motion Imitation for Learning Cart Pushing with Legged Manipulators
Mar 27, 2026Mili Das, Morgan Byrd, Donghoon Baek +1Adversarial Motion Prior Robotic Loco-Manipulation
Georgia Institute of Technology, Atlanta, GA, 30308, USA

Oct 7, 2026J/K move · Enter open · S save

HuMBLE: Human Motion-Driven Behavior Learning for Embodied Locomotion

Mike Zhang, Dongho Kang, Kevin Bergamin, Nicola Burger+18

RAI Institute, USA · Boston Dynamics, USA · Carnegie Mellon University, USA

Human-to-Robot Motion Imitation Humanoid Robot Locomotion RL for Legged Locomotion Robot Imitation Learning

Open paper AskPDF arXiv

Despite recent advances in humanoid locomotion, controllers optimized for command tracking and robustness tend to produce mechanical gaits, whereas controllers tied to human motion data often fail to generalize to commands outside the data distribution. This work introduces a learning framework that balances these competing objectives to synthesize real-time steerable, robust, and biomimetic locomotion policies from human data. Using an in-house curated locomotion dataset covering diverse speeds and directions, we first learn a natural locomotion prior policy through a teacher-student distillation process. Specifically, we train a full-body reference-conditioned policy with Reinforcement Learning (RL), then distill it into a lightweight prior policy conditioned solely on proprioception and a planar torso-velocity steering command. Next, we fine-tune the prior policy with multi-task RL to expand command coverage and robustness beyond the data distribution, pairing a goal-conditioned task that tracks arbitrary commands with a reference-guided task that tracks the human data as an explicit style regularizer. We validate our framework on three humanoid robots: the Boston Dynamics Atlas R1, Atlas D1, and Unitree G1. Experimental results demonstrate robust performance across real-world scenarios, including direct user-controlled locomotion in indoor and outdoor environments, and integration as the locomotion layer within hierarchical control stacks. Benchmarks against Tabula Rasa RL policies trained without human data and ablation studies confirm that our framework yields a lightweight, deployable policy that reconstructs coordinated whole-body behavior from a steering command, retaining the human gait characteristics while remaining robust and fully steerable.

Abstract

Humanoid Robot Locomotion

Explore similar work

HuMBLE: Human Motion-Driven Behavior Learning for Embodied Locomotion

Natural Sit-to-Stand Motion Synthesis For Humanoids via Guided Assistance Curricula and Staged Rewards

Partial Motion Imitation for Learning Cart Pushing with Legged Manipulators