cs.ROOct 17, 2026

Action-Aligned Latent Planning for Versatile Humanoid Loco-Manipulation

Authors: Yutang Lin, Guanya Shi, C. Karen Liu

Organizations: Amazon FAR. · Carnegie Mellon University. · Stanford University.

Read PDF
Ask AI
Cite

Copy citation

DOINot available for this paper

Abstract

Learning a single perceptive policy for diverse humanoid loco-manipulation requires unifying high-level task decisions without losing the state-dependent physical responses that make each interaction succeed. This makes the representation between high-level planning and whole-body control critical: kinematic or contact targets provide compact abstractions but do not directly encode interaction-specific control responses, while directly predicting low-level actions requires the planner to model high-frequency closed-loop feedback. We propose FIBER, a framework that uses the executable latent space of a frozen behavioral foundation model (BFM) as a shared planning representation for heterogeneous interaction skills. Our latent policy projection (LPP) maps independently acquired task-specific policies into this representation by identifying latent commands whose induced actions match their local control responses, enabling lower-rate planning while retaining the BFM's high-frequency state feedback. After projection, we aggregate closed-loop latent rollouts from all skills and train a single perceptive flow planner entirely offline. We further introduce AH-BFM, an adaptive-horizon BFM trained with multi-gamma forward-backward (FB) learning to support both precise control and temporally extended whole-body behavior. We evaluate FIBER on box pickup, door opening, and object grasping in simulation and on a real Unitree G1 humanoid. Across 128 diverse variations per task in simulation, FIBER achieves 42.0% mean success, more than tripling projected latent replay (11.6%) and substantially outperforming kinematic planning (3.5%). Real-world experiments further demonstrate sim-to-real transfer, skill chaining, and conditional steering through the same unified planner.

Figures & tables

Fig. 1 : Overview of Fiber . (a) Heterogeneous interaction experts are acquired through motion imitation, reward shaping, or successor-feature-guided behavior reuse. (b) Action-aligned latent policy projection maps their short-horizon control responses into a frozen adaptive-horizon bfm ; closed-loop rollouts produce executable latent trajectories. (c) These trajectories train a unified perceptive planner entirely offline. Language commands or waypoint conditions are encoded as steering tokens, and a DiT planner predicts future latent trajectories. During deployment, the planner replans from onboard observations while the frozen bfm provides high-frequency feedback.
MethodPickup BoxOpen DoorGrasp ObjectMean
w/ appr.w/o appr.w/ appr.w/o appr.
Task-specific policies
Expert Policy †100.0 ±\pm 0.090.8 ±\pm 0.589.3 ±\pm 0.297.9 ±\pm 0.897.0 ±\pm 0.895.4
lpp w/ SONIC [ 34 ] †0.0 ±\pm 0.00.0 ±\pm 0.083.1 ±\pm 0.788.1 ±\pm 0.65.1 ±\pm 0.729.4
LPP w/ Fixed- γ\gamma BFM †75.4 ±\pm 0.280.2 ±\pm 1.361.8 ±\pm 1.972.4 ±\pm 0.769.3 ±\pm 1.268.8
lpp w/ AH-BFM †81.9 ±\pm 0.987.4 ±\pm 2.483.5 ±\pm 1.897.3 ±\pm 0.452.9 ±\pm 2.172.8
TABLE I : Simulation success rates (%) across heterogeneous loco-manipulation tasks. “w/ appr.” includes the approaching phase, whereas “w/o appr.” initializes the humanoid near the interaction state. † denotes a task-specific policy, and ∗ denotes access to privileged object states instead of egocentric observations. Mean is computed using the “w/ appr.” results.
Fig. 2 : Skill chaining with Fiber . In a single rollout, the policy sequentially opens the door in (i), picks up the box in (ii), and carries it back through the doorway in (iii), demonstrating closed-loop chaining of multiple interaction skills under changing high-level commands.
Fig. 3 : Qualitative real-world behaviors of Fiber . (a) During box carrying, the robot reacts to an unstable grasp by briefly lifting the box and repositioning its hands underneath for more secure support, illustrating high-frequency closed-loop adjustment. (b) After picking up the box, the same policy follows waypoint commands while carrying it; after stumbling, the robot stabilizes on its knees and recovers before continuing, demonstrating interaction-preserving skill chaining, steering, and recovery. (c) For a heavy door, the robot leans its whole body into the interaction to generate sufficient pushing force, showing adaptive whole-body realization of contact-rich tasks.
MethodPickup BoxOpen DoorGrasp ObjectMean
Fiber6/104/102/1040%
TABLE II : Real-world success rates of the unified Fiber planner on the Unitree G1.
MethodJoint-MAE ↓\downarrowBE-EMD ↓\downarrow
0.2 s0.5 s1 s2 s
SONIC [ 34 ]0.1315––––
BFM γ0.8\gamma_{0.8}0.11070.0380.2000.4950.792
BFM γ0.98\gamma_{0.98}0.13270.0480.2130.4510.601
AH-BFM0.10200.0490.1950.4220.590
TABLE III : Ablation of multi-horizon BFM tracking and behavior embedding performance on 100STYLE [ 36 ] .
Appendix figures & tables7 assets

Supplementary material from the paper’s appendix.

Appendix

ComponentHidden widthHidden layers
Body actor20486
Forward representation20486
Critics20486
Backward encoder10241
Motion discriminator10243
Executable latent dimension256
Appendix
TABLE IV : BFM architecture used in our implementation.
ParameterValue
Forward-map learning rate3×10−43\times 10^{-4}
Backward-map learning rate10−510^{-5}
Actor / critic learning rate3×10−43\times 10^{-4}
Discriminator learning rate10−510^{-5}
Minibatch size1024
Training updates100,000
Appendix
TABLE V : BFM optimization parameters. Target networks use Polyak updates θˉ←(1−ρ)θˉ+ρθ\bar{\theta}\leftarrow(1-\rho)\bar{\theta}+\rho\theta .
ParameterMotion expertSF adapter
Rollout steps / environment3232
Learning rate3 ⁣× ⁣10−43\!\times\!10^{-4}10−510^{-5}
Optimization epochs35
Minibatches / rollout84
PPO clipping parameter0.20.2
Discount0.990.99
Appendix
TABLE VI : PPO parameters for the motion expert and the SF-guided composition adapter.
ParameterValue
OptimizerAdamW
Learning rate3×10−43\times 10^{-4}
Weight decay10−410^{-4}
Gradient-norm limit1
Minibatches / rollout4
Optimization passes / rollout5
Appendix
TABLE VII : Latent-projector optimization parameters.
Component or parameterConfigurationComponent or parameterConfiguration
Executable body latentdz=256d_{z}=256Transformer width / latent blocks1024 / 16
Prediction horizon11 sAttention heads / per-head width8 / 128
Body / optional hand state58 / 14Feed-forward expansion / activation4 / GELU
Optional hand target14 per latent stepAttention / feed-forward dropout0 / 0.2
Depth image / patch shape144×256144\times 256 / 12×1612\times 16Conditioning registers8
Depth range / spatial tokens0.120.12 – 3.03.0 m / 192Flow-time embedding scale1000
Appendix
TABLE VIII : Architecture of the unified latent flow planner.
ParameterValue
OptimizerAdamW
Learning rate2×10−42\times 10^{-4}
Weight decay10−210^{-2}
Warm-up updates100
Post-warm-up scheduleConstant
Batch size256
Appendix
TABLE IX : Offline planner optimization parameters.
TaskVariationEvaluation range
Box pickupBox geometry, placement, and massBase size 20×30×4020\times 30\times 40 cm, independently scaled by sx,sy∈[0.5,1.5]s_{x},s_{y}\in[0.5,1.5] and sz∈[0.5,1.0]s_{z}\in[0.5,1.0] , yielding sizes from 10×15×2010\times 15\times 20 cm to 30×45×4030\times 45\times 40 cm. Boxes are placed either on the ground or on supports 0.250.25 – 0.500.50 m high, with mass sampled from 0.50.5 – 1.51.5 kg.
Door openingDoor geometry and handle placementDoor width 0.80.8 – 1.11.1 m and height 1.91.9 – 2.22.2 m. Handle height is 0.850.85 – 0.950.95 m above the ground, with lateral inset 0.080.08 – 0.150.15 m.
Object graspingObject identity and shape128128 distinct object instances are evaluated with four takes per object ( 512512 episodes total), selected through teacher-feasibility screening from 685685 object instances and 2,9912{,}991 usable demonstrations.
Appendix
TABLE X : Simulation evaluation variations.
More in this topic

Humanoid Loco-Manipulation

Humanoid loco-manipulation studies how robots coordinate walking, balance, and whole-body interaction to move and handle objects.

See all 86 papers

Explore similar work

CardsList
Aug 7, 2026

LUCID: Latent-Skill Unified Control via Imagined Dynamics for Long-Horizon Humanoid Loco-Manipulation

Long-horizon humanoid loco-manipulation requires composing versatile whole-body skills and reliable high-level decision making. Existing methods often coordinate pretrained skills with scripted planners, finite-state machines or task-specific model-free policies, restricting their ability to handle complex task sequences. To address this limitation, we propose LUCID, a hierarchical model-based reinforcement learning framework that plans over reusable skills through imagined rollouts of a learned dynamics model. LUCID first trains a structured latent-conditioned low-level policy via adversarial imitation and then freezes it while jointly learning a high-level policy and macro-dynamics world model. The world model predicts the temporally extended state transitions induced by latent decisions, enabling high-level policy optimization through imagined rollouts. We evaluate our framework across various simulated multi-object rearrangement scenarios. Experimental results show that LUCID improves the full-task success and partial-completion rates compared to prior baseline methods, demonstrating its effectiveness in complex sequential loco-manipulation tasks.

Cheng Guo, Mingzhe Ni, Angelo Cangelosi +1

Department of Computer Science, University of Manchester, Manchester, UK · Human-Robot Interfaces and Interaction Laboratory, Italian Institute of Technology, Genoa, Italy

Humanoid Loco-ManipulationLatent Action LearningRobot Skill LearningModel-Based RL
AskPDF
Oct 17, 2026

DiPLO: Self-distilled Diffusion Planning for Loco-manipulation

Humanoid loco-manipulation is particularly challenging because it requires coordinating locomotion, balance, and object interaction, making it difficult to learn from limited demonstrations or from scratch through Reinforcement Learning (RL). We introduce DiPLO (self-distilled Diffusion Planning for Loco-manipulation), a framework that combines a self-distilled diffusion planner with a low-level whole-body tracking policy. Our planner is pretrained via supervised learning on motions derived from human-to-humanoid retargeting, while the tracker is trained via RL to execute them. DiPLO centers around two complementary post-training stages. First, we introduce Self-Distillation Diffusion Policy Optimisation (SD:DPO) to remove the planner's dependence on strong task conditioning, which we identify as a crucial ingredient for effective pretraining, as downstream fine-tuning does not recover comparable performance when it is omitted.. We then fine-tune this distilled model with Diffusion-Augmented Proximal Policy Optimization (DA:PPO), a maximum-entropy RL method for diffusion models, to improve task success and align generated plans with the frozen tracker's execution capabilities. Together, these stages enable motion generation through state-history and goal conditioning alone, enabling 99% task success in pick-carry-place tasks, compared to just 45% for the pretrained model. The resulting planner also generates motions for unseen object goals, with further evaluations on kicking and pushing/pulling tasks. Finally, we deploy our pipeline on the hardware, using our DA:PPO planner-tracker on a Unitree G1, achieving repeated successful executions without any real-world fine-tuning.

Shafeef Omar, Sebastian Sanokowski, Elham Daneshmand +4

Technical University of Munich (TUM), Munich, Germany. · Institute for Advanced Study (IAS), Technical University of Munich, Munich, Germany. · McGill University, Montreal, Canada. +1

Humanoid Loco-ManipulationDiffusion-Based PlanningRL for RoboticsSelf-Distillation
AskPDF
Oct 5, 2026

I-BFM: Reward-Conditioned Robust Humanoid Interaction via Unsupervised Reinforcement Learning

Behavioral foundation models (BFMs) have recently shown that a single humanoid policy can support diverse whole-body control, but extending such generality to physical interaction remains challenging. We introduce I-BFM, to our knowledge the first BFM for humanoid-object interaction. Rather than relying on task-specific policies or reference tracking, I-BFM learns a shared representation of the coupled dynamics among the humanoid, objects, and their contacts using forward-backward representations and unsupervised reinforcement learning. Given a downstream task reward, the same policy can be directly conditioned on a latent command to execute closed-loop interaction without task-specific policy optimization. To improve interaction control over different time scales, we further train the policy with both short-horizon interaction targets and longer-horizon goal targets. A single I-BFM policy performs carrying, pushing, and kicking, while also supporting goal reaching, motion tracking, stylistic control, and long-horizon task chaining. More importantly, it remains effective after large deviations from nominal execution: on Carry, I-BFM achieves 94.3% nominal success and retains 89.3% success after robot falls, compared with 1.3% for a planning-based baseline. Real-world experiments on a Unitree G1 further demonstrate diverse loco-manipulation behaviors, rapid recovery from interaction failures and external disturbances, and task chaining without task-specific retraining.

Ziqi Han, Yitang Li, Junhan Sun +8

Tongji University · Tsinghua University · Zhejiang University +3

Representation LearningReinforcement LearningHumanoid Robot ControlUnsupervised RLRobust RLHuman-Object Interaction
AskPDF