Publications.

Learning agile humanoid behaviors from human motion offers a powerful route to natural, coordinated control, but existing approaches face a persistent trade-off: reference tracking policies are often brittle outside the demonstration dataset, while purely task-driven Reinforcement Learning (RL) can achieve adaptability at the cost of motion quality. We intro duce a unified multi-task RL framework that bridges this gap by treating reference motion as a prior for behavioral shaping rather than a deployment-time constraint. A single goal-conditioned policy is trained jointly on two tasks that share the same observation and action spaces, but differ in their initialization schemes, command spaces, and reward structures: (i) a reference guided imitation task in which reference trajectories define dense imitation rewards but are not provided as policy inputs, and (ii) a goal-conditioned generalization task in which goals are sampled independently of any reference and where rewards reflect only task success. By co-optimizing these objectives within a shared formulation, the policy acquires structured, human-like motor skills from dense reference supervision while learning to adapt these skills to novel goals and initial conditions. This is achieved without adversarial objectives, explicit trajectory tracking, phase variables, or reference-dependent inference. We evaluate the method in a challenging box-based parkour playground that demands diverse athletic behaviors (e.g., jumping and climbing), and show that the learned controller transfers beyond the ref erence distribution while preserving motion naturalness. Finally, we demonstrate long-horizon behavior generation by composing multiple learned skills, illustrating the flexibility of the learned polices in complex scenarios.

Achieving robust, human-like whole-body control on humanoid robots for agile, contact-rich behaviors remains a central challenge, demanding heavy per-skill engineering and a brittle process of tuning controllers. We introduce ZEST (Zero-shot Embodied Skill Transfer), a streamlined motion-imitation framework that trains policies via reinforcement learning from diverse sources—high-fidelity motion capture, noisy monocular video, and non physics-constrained animation—and deploys them to hardware zero-shot. ZEST generalizes across behaviors and platforms while avoiding contact labels, reference or observation windows, state estimators, and extensive reward shaping. Its training pipeline combines adaptive sampling, which focuses training on difficult motion segments, and an automatic curriculum using a model-based assistive wrench, together enabling dynamic, long horizon maneuvers. We further provide a procedure for selecting joint-level gains from approximate analytical armature values for closed-chain actuators, along with a refined model of actuators. Trained entirely in simulation with moderate domain randomization, ZEST demonstrates remarkable generality. On Boston Dynamics’ Atlas humanoid, ZEST learns dynamic, multi-contact skills (e.g., army crawl, breakdancing) from motion capture. It transfers expressive dance and scene-interaction skills, such as box-climbing, directly from videos to Atlas and the Unitree G11. Furthermore, it extends across morphologies to the Spot quadruped, enabling acrobatics, such as a continuous backflip, through animation. Together, these results demonstrate robust zero-shot deployment across heterogeneous data sources and embodiments, establishing ZEST as a scalable interface between biological movements and their robotic counterparts.

This paper extends a previously proposed fall prediction algorithm to a real-time (online) setting, with implementations in both hardware and simulation. The system is validated on the full-sized bipedal robot Digit, where the real-time version achieves performance comparable to the offline implementation while maintaining a zero false positive rate, an average lead time (defined as the difference between the true and predicted fall time) of 1.1s (well above the required minimum of 0.2s), and a maximum lead time error of just 0.03s. It also achieves a high recovery rate of 0.97, demonstrating its effectiveness in real-world deployment. In addition to the real-time implementation, this work identifies key limitations of the original algorithm, particularly under omnidirectional faults, and introduces a fine-tuned strategy to improve robustness. The enhanced algorithm shows measurable improvements across all evaluated metrics, including a 0.05 reduction in average false positive rate and a 1.19s decrease in the maximum error of the average predicted lead time.

This paper presents a novel approach to fall prediction for bipedal robots, specifically targeting the detection of potential falls while standing caused by abrupt, incipient, and intermittent faults. Leveraging a 1D convolutional neural network (CNN), our method aims to maximize lead time for fall prediction while minimizing false positive rates. The proposed algorithm uniquely integrates the detection of various fault types and estimates the lead time for potential falls. Our contributions include the development of an algorithm capable of detecting abrupt, incipient, and intermittent faults in full-sized robots, its implementation using both simulation and hardware data for a humanoid robot, and a method for estimating lead time. Evaluation metrics, including false positive rate, lead time, and response time, demonstrate the efficacy of our approach. Particularly, our model achieves impressive lead times and response times across different fault scenarios with a false positive rate of 0. The findings of this study hold significant implications for enhancing the safety and reliability of bipedal robotic systems.

This dissertation focuses on developing a fall-tolerant framework for bipedal robots, aiming to enhance their ability to navigate challenging situations by effectively assessing, adapting, and responding to uncertainties and disturbances. Bipedal robots, with their unique capability to navigate diverse terrains and restore mobility, are ideal for assisting in critical and day-to-day tasks. However, their real-world deployment is limited due to factors like high-dimensional complex dynamics and a smaller support polygon,making it difficult to achieve stable motion, especially in the face of disturbances and uncertainties.

To address these limitations, the dissertation develops robust controllers and reliable fall prediction algorithms. Feedback controllers have been used in the literature to ensure robustness against disturbances and uncertainties. However, the infeasibility of accounting for all disturbances and uncertainties during real-world operations makes falls inevitable. Falls are undesirable as they can prevent a robot from completing its task, result in damage to the surrounding area, or lead to injuries. Therefore, the dissertation emphasizes the importance of implementing robust controllers and employing methods to predict falls.

This research begins by introducing a systematic method to design control objectives for highly constrained systems and concludes by presenting a 1D convolutional neural network fall prediction algorithm capable of not only predicting falls but also estimating the time to react. The effectiveness of the control objectives is demonstrated through robust, comfortable closed-loop sit-to-stand motions for a fully actuated lower limb exoskeleton, Atalante. The performance of the proposed fall prediction algorithms is evaluated in simulation using a planar-four link robot based on Atalante and in hardware and simulation for the bipedal robot Digit.

For legged robots to operate in complex terrains, they must be robust to the disturbances and uncertainties they encounter. This paper contributes to enhancing robustness through the design of fall detection/prediction algorithms that will provide sufficient lead time for corrective motions to be taken. Falls can be caused by abrupt (fast-acting), incipient (slow-acting), or intermittent (non-continuous) faults. Early fall detection is a challenging task due to the masking effects of controllers (through their disturbance attenuation actions), the inverse relationship between lead time and false positive rates, and the temporal behavior of the faults/underlying factors. In this paper, we propose a fall detection algorithm that is capable of detecting both incipient and abrupt faults while maximizing lead time and meeting desired thresholds on the false positive and negative rates.

Lower-limb exoskeletons provide people who suffer from lower limb impairments with an opportunity to stand up and ambulate. Standing up is a crucial task for lower-limb exoskeletons as it allows the user to transfer to the exoskeleton from a wheelchair, with no assistance, and can be a precursor to walking. Achieving a safe sit-to-stand motion for the exoskeleton + user system can be challenging because of the need to balance user comfort while respecting hardware bounds and being robust to changes in the user characteristics and the user's environment. We successfully achieve safe sit-to-stand motions by using constrained optimization to generate two types of dynamic sit-to-stand motions based on two hybrid system descriptions for the exoskeleton, Atalante. Due to the highly constrained nature of the equations of motions, we introduce a method to systematically design virtual constraints for highly constrained systems. We also design two quadratic program-based computed-torque controllers to achieve the sit-to-stand motion and to safely come to a stop in a standing position. We then analyze the closed-loop behaviors of the two sit-to-stand motions under the two controllers using physically motivated robustness tests. The criteria used to determine a successful sit-to-stand motion are: tracking error, the pitch acceleration of the torso, the amount of user force needed to perform the motion, and the adherence to the Zero Moment Point (ZMP), friction, and joint constraints.