The Poppy Humanoid is an open-source, low-cost robot suitable for research and education in artificial intelligence. However, we are unaware of any published methodology that achieves reliable, unassisted bipedal locomotion on the standard Poppy hardware. This paper contributes a functional closed-loop walking controller for Poppy, based on the linear-quadratic regulator (LQR) framework for trajectory tracking. Starting with data collected from open-loop playback of a nominal walking trajectory, our proposed method learns a quadratic cost function for an LQR controller that substantially improves the reliability of the motion. The closed-loop controller is validated empirically, demonstrating statistically significant improvements in walking performance compared to open-loop trajectory playback.
Figures & tables
Figure 1: The Poppy hardware and four locations used in our experiments. To prevent hardware damage during falls, we fashioned an experimental harness from a consumer clothing rack and camera strap, with a custom wooden insert to avoid pressure on the sides of the head when the strap is taut. We only count footsteps as “successful” if the strap is slack for the entirety of the motion without any human intervention or assistance.
Figure 2: Nominal trajectory (solid lines) and recorded joint angles (dashed lines) from a representative hardware run. Circles indicate the timing of the waypoint commands. Only a subset of joint motors are shown for figure legibility, labeled at right according to the standard Poppy naming convention.
Figure 3: Histogram over open-loop runs of the number of successful footsteps before a fall (trajectories were 6 footsteps long, so 6 successful footsteps means no fall during the episode). The left panel is for noiseless trajectories only ( σ=0.0 , see text); the right panel is over all runs.
Figure 4: Per-run, per-timepoint Euclidean distances from current joint observations to average successful run (left) or initial stance pose before each footstep cycle (right). On right we only show the timepoints at the boundary of each cycle. Each transparent curve is a separate run, and opaque curves show the means over all runs in their respective categories. Data from failure runs after the recorded fall step are omitted from the plots.
Figure 5: Left: Mean absolute deviation between dynamics model predictions and observed data. Right: Condition number of fitted dynamics matrix.
Figure 6: Left : Cumulative products Λn of maximum eigenvalue magnitudes of An+BnKN , for several ϵ . “Open-loop” refers to “passive” residual dynamics with no LQR adjustments, i.e., using maximum eigenvalues of An . Middle: Eigenvalues of the cost coefficient matrices Cn . Each scatterpoint is an individual eigenvalue and solid lines show minimum eigenvalue for each ϵ and n . Right: Violin plots showing cost distributions over failed and successful runs in the open-loop data for ϵ=−2.0 (each scatterpoint is a run with horizontal noise added for legibility).
Quadruped locomotion provides a natural setting for understanding when model-free learning can outperform model-based control design, by exploiting data patterns to bypass the difficulty of optimizing over discrete contacts and the combinatorial explosion of mode changes. We give a principled analysis of why imitation learning with quadrupeds can be inherently effective in a small data regime, based on the structure of its limit cycles, Poincaré return maps, and local numerical properties of neural networks. The understanding motivates a new imitation learning method that regulates the alignment between variations in a latent space and those over the output actions. Hardware experiments confirm that a few seconds of demonstration is sufficient to train various locomotion policies from scratch entirely offline with reasonable robustness.
Bipedal loco-manipulation enables robots to interact with objects beyond the nominal workspace of their arms by coordinating locomotion and manipulation. Realizing this capability requires a low-level whole-body controller that translates task-level manipulation goals into coordinated arm and leg motions while maintaining balance. We present a unified whole-body controller trained with reinforcement learning that directly maps 6-DoF end-effector targets to coordinated actions for the bipedal base and robotic arm. Given only an end-effector target, the learned controller autonomously coordinates reaching, postural adaptation, and stepping without explicit base-velocity or footstep commands. A reward-gating strategy regulates the trade-offs among end-effector tracking, locomotion, and balance during training, while a temporal context estimator combines windowed Transformer encoding, recurrent GRU memory, and auxiliary dynamics prediction to extract dynamics-relevant information from observation history. Real-robot experiments demonstrate that the same controller supports reaching, postural adaptation, and stepping under commands from VR teleoperation, a learned diffusion policy, and scripted trajectories, providing a common end-effector interface for diverse manipulation tasks.
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.
Mike Zhang, Dongho Kang, Kevin Bergamin +19
RAI Institute, USA · Boston Dynamics, USA · Carnegie Mellon University, USA