SBIR-STTR Award

Human-worn powered exoskeleton robots can provide extensive benefits to the USAF, the DoD, as well as the commercial and industrial sectors in efficiently and effectively performing tasks such as: loading and unloading boxes/crates and equipment; loading munitions; assembling pallets; and in many other applications. The benefit of exoskeleton robots can be multiplied by allowing these to learn from their operators how to execute tasks and then allowing these robots to perform these tasks on their own operating under supervised autonomy (man-on-the-loop). Phase I of this effort aims to assess the applicability of physics-based simulation and Machine Learning (ML) tools tailored to rapidly allow exoskeleton robots to learn, from their operators, how to perform a collection of USAF relevant tasks, and how to integrate this new capability into the customers and end-users work flow.

Human-worn powered exoskeleton robots can provide extensive benefits to the USAF, the DoD, as well as the commercial and industrial sectors for efficiently and effectively performing tasks such as: loading and unloading boxes/crates and equipment; loading munitions; assembling pallets; and in many other applications. The benefit of exoskeleton robots can be multiplied by allowing them to learn from their operators how to execute tasks and then allowing these robots to perform these tasks on their own or as cobots, operating under supervised autonomy (man-on-the-loop). Under this Phase II effort, the team from Sarcos will extend the capabilities of its human operated powered exoskeleton (the XO® system) to allow human-guided success-based training of physics-based models of the CytarTM upper extremity (i.e., a pair of dexterous arms) to perform manipulation-intensive tasks that are relevant to USAF customers and end-users. By combining human inputs as the starting point, and machine learning methods to perform multiple simulations concurrently while introducing variations in robot kinematics and dynamics, as well as sensors, environment, and object variability, a robust database of control trajectories will, quickly, and at relatively low cost, be generated and then be used to perform similar tasks in the presence of real-world uncertainties.