mihailp @ seas.upenn.edu   +1-412-805-0210            Curriculum Vitae

Robot Navigation

A number of robot applications require efficient  and reliable obstacle avoidance and goal acquisition in difficult, cluttered and uncertain environments. Robot performance is often measured in terms of time to traverse given terrain, energy expenditure, and other qualities. Efficient motion planning in this setting is one of the challenges that makes this problem hard. Many fielded solutions feature two-tiered planning solutions consisting of a "global" planner that provides guidance toward the goal (and trades off quality for efficiency over long-ranges) and a "local" planner that computes immediate actions (gains quality at the cost of short-range reasoning). In this line of work, we focus on the  "local" component and develop robot mobility representations of higher quality than utilized in standard practice. We also research improving "global" components in our constrained planning work.

Acknowledgements: NASA GSRP Fellowship, JPL SURP
   NASA logo JPL logo

Path Relaxation

path relaxation
Most planning approaches gain runtime efficiency via discretized representations. However, the rigidity of fixed representation leads to inevitable drawbacks in planner completeness. This line of work applies optimization techniques to gain flexibility in representation and, therefore, generate a better approximation to the contiuum of the system's reachability.

Clothoid Parameterizations

Many fielded navigation solutions feature local planners that evaluate a set of constant-curvature arcs. We explored applying clothoids, linear curvature functions of path length. The additional degree of freedom decouples end-point position from orientation and allows "S"-shaped motions. Improved performance was demonstrated at no additional cost in computation.

See publications on this topic.