Researchers' MPC framework lets robots autonomously open and traverse doors

Mobile manipulators in human spaces routinely face an everyday challenge: doors. A new paper on arXiv (2608.00206) from researchers Kasra Sinaei, Kasun Weerakoon, Christopher Bradley, Seyed Abolfazl Fakoorian, and Donald Ebeigbe presents a motion planning framework that treats the robot and the door as a single coupled dynamical system within a nonlinear Model Predictive Control (MPC) optimization. This unified approach lets the planner generate dynamically feasible, collision-free trajectories for both pushing and pulling doors, rather than relying on separate base and arm controllers. A notable design choice is enforcing manipulation feasibility through a penalty-based constraint, which avoids the complexity and computational burden of explicitly modeling the arm's kinematics in the planner.

The team validated their approach through simulations and a hardware experiment, demonstrating that it can successfully plan and execute feasible trajectories for door traversal. The results suggest that this method could make autonomous navigation more robust and practical for mobile manipulators in offices, hospitals, and homes, where doorways are the norm. By eliminating the need for detailed arm models during planning, the framework may also generalize more easily to different robot platforms and reduce engineering effort. While the paper doesn't specify speed or accuracy numbers, the key advance is treating the door as part of the robot's dynamics, enabling smoother and more reliable interaction with a common but tricky obstacle. This work adds a practical stepping stone toward fully autonomous mobile manipulation in human-centric environments, and it's a promising direction for real-world deployment of service and industrial robots.