Behavior Tree Based Reinforcement Learning for Robotic Arm Task Planning
Robotic arms are typically programmed to execute motion cycles through conventional programming paradigms. The burden is on the programmer to provide for all scenarios however infrequent unexpected events may occur. Behavior trees offer a structured approach to model the expected behavior of the arm as a hierarchical tree with nodes that can handle the flow of execution to achieve a desired outcome. Behavior trees also lend themselves to reinforcement learning methods to refine and extend the behavior based on data collected during use of the robotic arm.
The intern will get hands-on experience with the full cycle of developing the learning feature on a robotic arm in the company’s product portfolio and testing it in a real world scenario.
Bachelor’s or Master’s degree student from the relevant field
Familiarity with two or more of:
-Kinematics and dynamics of 6 dof serial link robotic arms
-Variational calculus and optimization methods
-Fundamentals of behavior trees