Last updated
Charles Noren, Burhanuddin Shirose, Bhaskar Vundurthy, Sebastian Scherer, and Matthew Travers
During off-road operations, mobile robotic platforms often encounter objects that influence the platform’s route. As determining the outcome of an interaction with an object (e.g., overriding) is difficult, many robotic planners are designed to avoid interactions with all environmental objects. Yet, this object-adverse planning behavior is not reflected in the actions of expert human operators, who may interact with objects in order to find a viable path towards their goal. This work intends to emulate that human operator intuition. Our objective is to improve the performance of robotic traversals in off-road terrains through the development of a planning paradigm that allows certain safe contact with environmental objects. Specifically, we design a two-level hierarchical planning and control system which couples a contact-informed regional motion planner with contact-constrained local nonlinear trajectory optimization techniques. The approach is demonstrated for classes of vegetative objects which are characterized by existing parameterized collision models from the terramechanics community. The top level of the hierarchy combines sampling-based planning techniques with a set of override (velocity) constraints derived from these collision models during the search for a minimum-time trajectory towards the platform’s goal. This minimum-time trajectory is then passed to the lower-level of the architecture, which utilizes direct collocation and model predictive control techniques to ensure that the velocity constraints are enforced during trajectory execution. The capabilities of the architecture are shown both in simulation and onboard a robotic platform, where vegetation is reasoned about and then overridden depending on the environment, platform, and object geometry.
Full paper purchase: ISTVS members receive three complimentary papers per year: