Safety and stability are common control requirements for affine-control systems. To check if a potential controller is feasible in ensuring stability and safety, the control Lyapunov function(CLF) and the control Barrier function(CBF) are usually used respectively. However, it is still difficult to design a stable and safe controller for nonlinear and uncertain systems, such as controlling a stratospheric balloon within a wind field with reach-avoid objective. Hence, we are interested in investigating if a model-based learning approach (neural clbf) can synthesize robust and feasible feedback controllers for this underactuated control system. Results show that neural clbf can achieve a better performance in a simple wind-driven system, but more model design revisions are needed to be taken into considerations.

Reinforcement Learning (RL) algorithms aim to compute policies that maximize reward without necessarily guaranteeing safety during learning or execution. This makes it challenging to use RL in cyber-physical systems such as robotics, where unsafe actions may damage the robot and/or its environment. In contrast to most previous studies that propose techniques to reduce the number of unsafe actions given some prior knowledge of the environment, we are interested in the situation where undesired behaviors can be reduced without any prior system knowledge. Hence, in this study, we aim to explore a dynamic shielding technique that safeguards RL. Specifically, we implement a variant of RPNI, a type of automata learning, which is constructed in parallel with the model-based RL to filter out potential unsafe behaviors during action execution before the agent actually experiences it. Our experiment shows that the number of undesired trials can be effectively reduced compared to unshielded RL.

We propose a control synthesis framework for a heterogeneous multi-robot system to satisfy collaborative tasks, where actions may take varying duration of time to complete. We encode tasks using the discrete logic \( \text{LTL}^\psi \), which uses the concept of bindings to interleave robot actions and express information about relationship between specific task requirements and robot assignments. We present a synthesis approach to automatically generate a teaming assignment and corresponding discrete behavior that is correct-by-construction for continuous execution, while also implementing synchronization policies to ensure collaborative portions of the task are satisfied. We demonstrate our approach on a physical multi-robot system.

Robot caregiving tasks such as bathing, dressing, and transferring may require a robot arm to make contact with a human body at multiple points—not just at the robot's gripper or end effector. However, whole-arm contact presents challenges because varying human contact preferences may lead to unsafe or uncomfortable interactions. To address these challenges, we propose a novel algorithm that employs a conditional contextual bandit approach to distill user contact preferences into low-level pose and force control policies via a hierarchical operational space controller. Our approach intends to enable complex ma- nipulation tasks involving whole-arm contacts around humans, while adapting to contact preferences and maintaining control priorities. We propose a simulation-in-the-loop approach to minimize discomfort during preference learning, first gathering real-world feedback and then simulating further interactions in a digital twin environment. The robot refines its control policy iteratively in simulation, ensuring safe adaptation without direct experimentation on users until convergence. We perform a user study to develop a better understanding of contact preferences associated with physical robotic assistance and use the findings to initialize a realistic user model for our experiments. We validate our framework through a simulated bed-bathing task to demonstrate utility in caregiving. Results show that our framework effectively adapts to individual contact-related preferences while ensuring task completion and enhancing user comfort and safety during physical interactions.