Robot Control and Learning
Towards the better understanding of Behavior Trees and Motion Generators (BTMG) for Robot Behaviors
Faseeh Ahmad
The aim of this thesis is to evaluate the effectiveness of using Behavior Trees and Motion Generators (BTMG) policy representation for developing robotic skills to solve diverse tasks.
Our current approach involves utilizing SkiROS, a skill-based system, to create robotic skills. Each skill is divided into two parts: description and implementation. The skill description includes the parameters, pre- and post-conditions, while the implementation specifies the execution strategy. BTMG policy representation is used by SkiROS to represent the execution strategy of a skill. This parametric policy representation enables the specification of BT structure, control flow nodes, execution nodes, and the controller's actual stiffness values. In summary, BTMG policy provides details on both the execution strategy and how individual motions should be executed.
This study also aims to extend BTMG policy representation:
- Adapting extrinsic parameters of the BTMG representation to adapt to different variations of a task.
- Adapting intrinsic parameters of the BTMG representation to adapt to different tasks within similar problem description.
- Learning BTMG policy representation of a skill via Learning from demonstration and other state of the art methods