We are designing autonomous robotic exoskeletons to enhance human performance by assisting different tasks like walking, stair climbing, and sit-to-stand. These exoskeletons match human joint mechanics, prioritize lightweight design and comfort, and include hip and knee systems. An integrated AI co-processor enables real-time user state estimation for optimal biomechanical assistance.

We are developing stability and balance-aware exoskeleton control by integrating biomechanics, robotic control, and deep learning models. We analyze human movement to design exoskeletons that improve stability during dynamic tasks like ground perturbations. Advanced robotic control ensures precise assistance, while deep learning models enable real-time perturbation detections to assist the user.

We are working on a musculoskeletal model-based computer simulation using reinforcement learning to train exoskeleton control policies that can be zero-shot sim2real transferred. This approach optimizes control for diverse tasks and users without extensive human data collection. The simulated model ensures effective, real-world applicable control, enhancing mobility and reducing development time.

We are optimizing exoskeleton control parameters to enhance functional outcomes across varying tasks and users. By leveraging biomechanical model and control algorithms, we fine-tune assistance levels for both cyclic and non-cyclic lower-limb movements. Deep learning models adjust control profiles in real-time, ensuring optimal robot assistance that improves energy, stability, and user preference.

Personalizing exoskeleton control is critical in accommodating individual variances. We design deep learning models to automatically learn and adapt to novel user's gait using techniques like computer vision and physics simulation. This approach ensures real-time, tailored exoskeleton assistance for diverse user populations, with potential application to be used to clinical settings.

Exoskeleton controllers must generalize beyond constrained walking tasks, typically studied in a research lab. For this, we develop a task-generalizable exoskeleton control to enhance human mobility across diverse activities like walking, lunging, and jumping using deep learning, reducing joint effort and bridging the gap between lab-based exoskeleton technology and real-world mobility needs.