Abstract:
The navigation success of a wheeled mobile robotic mission is directly correlated to the degree of accuracy to which the robot can follow a given path. This, in turn, is largely affected by two factors: a) the environment and b) the intrinsic properties of the robot – its design, actuation mechanism etc.

In the first part of the talk, we focus on the impacts of the environment on the mobility of the robot in the form of wheel slip. In current practice, it is common to model the effects of wheel-slip as an exogenous disturbance and use feedback control to minimize tracking error. This approach breaks down when the robot operates in terrain that induces persistent wheel slip. We discuss an existing method to empirically model wheel slip and identify it. We then introduce a framework based on iLQR that uses the identified model to predictively compensate for wheel slip. We discuss the results of path following experiments in a data driven simulator and in the real world using our method and compare its performance to a baseline (pure pursuit) algorithm.

In the second part of the talk, we discuss one particular intrinsic property of the robot namely the actuator dynamics. We utilize a z-transform based model that captures the first order dynamics of the steering, brake and throttle actuators and compare its performance against an ad-hoc baseline actuator model. We then discuss an existing trajectory generation framework that predictively utilizes our model to generate steering controls that satisfy terminal position, heading (and its higher order constraints). We conclude by comparing the path tracking performance of this framework to a baseline algorithm (iterated pure pursuit) that uses the baseline actuator models in simulation for a real vehicle with significant actuation latencies.

Committee:
Alonzo Kelly
David Wettergreen
Cetin Mericli
Arun Srivatsan Rangaprasad