Integrating Symbolic Reasoning and Statistical Perception in Task and Motion Planning

A robot, like any cybernetic system, must perceive it’s world and act upon it to accomplish goals. Most goals are complex and must be broken down into simpler tasks. In the physical world, these tasks almost always require motion to accomplish. But motion itself is far from simple; joint movements must be coordinated, obstacles avoided, physical obstacles respected, and efficiencies maximized. Moreover, this task and motion planning problem is presented in a dynamic and incompletely-known environment. Solving this foundational problem requires high-level background knowledge about the world, understanding of the environment formed from sensory input, reasoning over this information, and low-level motion control for taking action in this environment.
In this project we focus on the integration of symbolic and statistical models for improving applications including perception, action selection, and execution in robots. This project aligns with one of the core research areas at Sanctuary AI, and allows us to further the work necessary to have robots actively participating in dynamic human environments.

Faculty Supervisor:

Kamal Gupta

Student:

Simon Odense

Partner:

Sanctuary AI

Discipline:

Engineering

Sector:

Information and cultural industries

University:

Simon Fraser University