Chuangchuang Sun
Postdoctoral Associate
Massachusetts Institute of Technology

The autonomy of robotics and space systems are fundamental issues, especially for large-scale systems with critical safety issues. Specific problems include robotic mixed-type decision-making problems and spacecraft multi-phase mission planning. We propose both off-line and online algorithms to empower autonomy in real-time.
We develop the off-line algorithm by first formulating such problems as a Quadratically Constrained Quadratic Programming (QCQP), with safety criterions as constraints directly. Subsequently, to solve the QCQP, an efficient optimization algorithm is proposed based on inexact augmented Lagrangian methods. Our algorithm admits simple subproblems with closed-form solutions, which leads to scalability and real-time applicability. Simulation results are presented to validate the effectiveness and efficacy of our algorithms.
Also, for the online algorithm, Control Barrier Functions (CBF) with forward-invariance is adopted to guarantee safety via calibrating the input from control algorithms. However, CBF in high-order systems can often encounter infeasibility due to control limitations. To address that, we learn a differentiable safety hyperplane getting the lower-order states involved. A feedback training scheme is developed to decrease the infeasibility rate recursively. Subsequently, the newly learned safety hyperplane is added as a constraint in the CBF formulation. Simulation results on path planning demonstrate the improvement of the proposed framework.

About the Speaker...

Dr. Chuangchuang Sun is currently a postdoctoral associate in the department of aeronautics and astronautics at MIT. He received his Ph.D. in August 2018 from the Ohio State University and a B.S. degree from the Beijing University of Aeronautics and Astronautics, China in 2013, both in Aerospace Engineering. His research interest focus on control, optimization, reinforcement learning and applications in robotics and space systems.