Humanoid robots deployed in industrial environments are required to perform load-carrying transportation tasks that tightly couple locomotion and manipulation. However, achieving stable and robust locomotion under varying payloads and upper-body motions is challenging due to dynamic coupling and partial observability. This paper presents a load-aware locomotion framework for industrial humanoids based on a decoupled yet coordinated loco-manipulation architecture. Lower-body locomotion is controlled via a reinforcement learning policy producing residual joint actions on kinematically derived nominal configurations. A kinematics-based locomotion reference with a height-conditioned joint-space offset guides learning, while a history-based state estimator infers base linear velocity and height and encodes residual load- and manipulation-induced disturbances in a compact latent representation. The framework is trained entirely in simulation and deployed on a full-size humanoid robot without fine-tuning. Simulation and real-world experiments demonstrate faster training, accurate height tracking, and stable loco-manipulation. Project page: https://lequn-f.github.io/LALO/
Figures & tables
Category
Dimension
Noise Scale
Phase Input
2
0
Projected gravity
3
±0.05
Angular velocity
3
±0.1
Joint positions
27
±0.02
Joint velocities
27
±0.5
Command
4
0
TABLE I: Observation space of the locomotion policy.
Fig. 1: Overview of the proposed load-aware humanoid loco-manipulation framework. Upper-body manipulation is generated by perception-driven kinematic control using 6D object pose estimation, while lower-body locomotion is governed by a residual reinforcement learning policy with height-conditioned offsets and history-based state estimation. Historical proprioceptive observations are encoded by a history-based state estimator to infer base velocity, base height, and a compact latent state capturing load- and manipulation-induced disturbances.
Reward term
Formulation
Weight (μi)
Linear velocity tracking
exp(−4∥[vx,vy]−[vx∗,vy∗]∥2)
1.2
Angular velocity tracking
exp(−4(ωz−ωz∗)2)
1.1
Base height tracking
exp(−50∣hb,rel−hb,rel∗∣)
2.0
Low-speed consistency
rls(vx,vx∗)⋅I(∣vx∗∣>0.1)
0.2
Orientation stabilization
exp(−10∥gproj,xy∥)
1.0
Velocity mismatch
0.5(exp(−10vz2)+exp(−5∥ωx,y∥))
0.5
TABLE II: Reward terms and weights used for training.
Parameter
Unit
Range
Operator
Ground Friction
–
[0.1, 3.0]
–
Restitution Coefficient
–
[0.0, 1]
–
External Push Velocity (XY)
m/s
[-0.5, 0.5]
additive
External Push Angular Velocity
rad/s
[0, 0.6]
additive
Base Mass Offset
kg
[-5.0, 10.0]
additive
Hand Mass Offset
kg
[-0.1, 0.3]
additive
TABLE III: Overview of Domain Randomization
Fig. 2: Multi-box detection and target 6D pose estimation. Blue boxes denote instances detected by YOLO-seg, the red box indicates the selected target, and the green wireframe shows the 6D pose estimated by FoundationPose.
Parameter Name
Value
Value loss coefficient
1.0
Number of Epochs
5
Clipping
0.2
Entropy Coefficient
0.003
Discount Factor
0.99
GAE Discount Factor
0.95
TABLE IV: PPO Training Hyperparameters
Fig. 3: Training convergence comparison of different offset and reference formulations. (a) Episode reward for tracking base height versus environment steps. (b) Mean episode length versus environment steps.
Fig. 4: Distribution of joint residual actions during steady-state walking.
Fig. 5: Prediction performance of the history-based state estimator. Comparison between commanded, ground-truth, and estimated states for (a) base velocity vx , (b) base velocity vy , and (c) base height. The estimator achieves low mean squared error under dynamic motion conditions.
Fig. 6: Real-world loco-manipulation experiments in an industrial box transportation scenario. (a–c) Box transportation without depalletizing: the robot transitions from unloaded walking to load-carrying locomotion, squats to a target height, places the box onto a rack while maintaining a crouched posture, and then stands up. (d–f) Transportation with depalletizing: the robot performs box depalletizing, walks while carrying a 5 kg payload, places the box at the designated location, and continues unloaded walking for the next task. The environment is a one-to-one reconstruction of an industrial factory setting and corresponds to the task scenario of the World Humanoid Robot Games.
Humanoid loco-manipulation of large, heavy objects demands forceful interaction across the entire body. However, such payloads shift a humanoid's center of mass and impose sustained loads across the upper body, challenging balance and command tracking. We present HULK, a whole-body control framework for forceful loco-manipulation. Using model predictive control (MPC) to guide reinforcement learning with predictions of the loaded dynamics, we train two teachers: one tracks arm motions under wrist forces, and the other locomotes while holding large objects against the body. A capture-point control barrier function augments the wrist-force teacher during training to improve balance under load. We distill both teachers into a single policy. Evaluation spans simulation and the Unitree G1. In simulation, the teacher with the barrier function achieves the lowest forward and lateral velocity tracking errors at 10 kg per arm among evaluated controllers and reduces aggregate divergent component of motion (DCM) excursion magnitude by 35.7% relative to MPC-guided reinforcement learning alone. Our wrist-force teacher withstands torso push disturbances of up to 130 N.
An Dang, Arturo Flores Alvarez, Yu-Ming Chen +5
Amazon · University of Michigan · University of California, Los Angeles +1
Humanoid robots could transport payloads substantially heavier than themselves by pulling passive wheeled vehicles instead of carrying the load. This capability, however, creates a coupled locomotion problem: the robot must maintain persistent upper-body contact while adapting to unknown, configuration-dependent forces arising from the payload, vehicle, and terrain. We present a whole-body control framework for humanoid rickshaw pulling that tracks commanded vehicle motion while preserving balance and stable grasps under uncertain load dynamics. During training, a privileged teacher exploits vehicle states, interaction forces, and load properties. Its actions and latent are distilled into a history-conditioned student that implicitly infers coupled dynamics from proprioceptive responses, followed by reinforcement-learning fine-tuning. Comparisons with No History and Only History baselines show that the resulting policy achieves accurate vehicle tracking while reducing vehicle oscillation, torso tilt, and actuation cost. Behavioral analysis shows that Unitree G1 propels the rickshaw and generates gait-synchronized whole-body reactions that stabilize its lateral and roll motions. Moreover, pulling redistributes joint effort and yields a lower robot-normalized cost-of-transport proxy than unloaded walking over most tested load--speed conditions. On hardware, a single policy performs starting, sustained pulling, turning, and stopping with both rigid payloads and human passengers, handling a loaded rickshaw mass of up to 115~kg without load-specific retuning. These results demonstrate robust heavy-load transportation through coordinated and persistent humanoid--vehicle interaction.
Most humanoid loco-manipulation controllers require human motion data to learn whole-body coordination and posture, leaving policies reliant on external sources to provide this data. We present OCLO (Online-posture Compliant LOco-manipulation), a humanoid loco-manipulation system trained without human motion data and commanded only through two end-effector targets. Because these targets do not uniquely determine whole-body posture, OCLO generates pelvis height and torso orientation online using an analytic reachability prior, further refined through policy-in-the-loop sampling with a task-agnostic cost. OCLO also learns whole-body compliance by displacing end-effector references according to measured forces through a spring-damper model, encouraging the legs, waist, and pelvis to yield to external loads. In simulation, using the reachability prior leads to a 77.8% success rate in acquiring the commanded reference, a vast improvement over the 37.8% success rate accomplished without the prior. Further, refinement reduces end-effector orientation error across all evaluated tasks. The same posture module improves a pretrained SONIC controller on four of five tasks. Without compliance training, policies tend to lose balance under disturbances rather than sacrifice tracking. On a Unitree G1, OCLO maintains balance under end-effector disturbances that cause its ablations to fail and performs seven loco-manipulation tasks, including crouched walking and picking up a box from a low surface. Project website: https://oclo-humanoid.github.io/
Seungho Yeom, Zhenyu Wu, Jaeyoung Huh +4
University of California San Diego, La Jolla, CA 92093, USA. · Yonsei University, Seoul 03722, Republic of Korea.