Register now After registration you will be able to apply for this opportunity online.
This opportunity is not published. No applications will be accepted.
Master Thesis: Hierarchical Optimal Motion Planning for Articulated Robots in Cluttered Environments
This thesis combines multiple layers of motion planners and optimal control algorithms in order to generate feasible and robust motion patterns for complex robots in non-standard environments.
Currently, when choosing a motion planning framework for a robotics task, one has essentially two options. Either choosing a 'sampling-based' motion planning algorithm or an Optimal Control based planner. The first class typically performs well in cluttered environments and can deal efficiently with obstacles, however does usually not provide optimal paths. (Some examples are shown in the appendix.) The second class obviously provides (locally) optimal solutions - however it assumes convexity and the optimization problem becomes ill-posed in cluttered environments, i.e. when many obstacles are present. This thesis project seeks to combine both approaches in a hierarchical fashion. While the high-level planning shall be performed with a sampling-based planner in order to create a problem of reduced size, an optimal control-based planner is to be applied for subsequent refinement of the solution.
Currently, when choosing a motion planning framework for a robotics task, one has essentially two options. Either choosing a 'sampling-based' motion planning algorithm or an Optimal Control based planner. The first class typically performs well in cluttered environments and can deal efficiently with obstacles, however does usually not provide optimal paths. (Some examples are shown in the appendix.) The second class obviously provides (locally) optimal solutions - however it assumes convexity and the optimization problem becomes ill-posed in cluttered environments, i.e. when many obstacles are present. This thesis project seeks to combine both approaches in a hierarchical fashion. While the high-level planning shall be performed with a sampling-based planner in order to create a problem of reduced size, an optimal control-based planner is to be applied for subsequent refinement of the solution.
With none of the currently existing approaches we can plan optimal trajectories for our robots in complicated settings (c.f. attached picture). The goal of this thesis is to combine existing motion planning building blocks into a new framework that allows to generate motions in a hierarchical way. You will integrate different planners such as RRT*, PRM*, STOMP and combine them with Sequential Linear Quadratic Optimal control and Multiple Shooting approaches. Finally, this will allow us to generate motion plans for different robots which avoid obstacles and fulfil local constraints at the same time.
With none of the currently existing approaches we can plan optimal trajectories for our robots in complicated settings (c.f. attached picture). The goal of this thesis is to combine existing motion planning building blocks into a new framework that allows to generate motions in a hierarchical way. You will integrate different planners such as RRT*, PRM*, STOMP and combine them with Sequential Linear Quadratic Optimal control and Multiple Shooting approaches. Finally, this will allow us to generate motion plans for different robots which avoid obstacles and fulfil local constraints at the same time.
- A master thesis which is itself quite "standalone" but with the potential for significant impact and improvement of the state of the art.
- A good working environment and the possibility to get in touch with state-of-the art robotics research.
- A master thesis which is itself quite "standalone" but with the potential for significant impact and improvement of the state of the art. - A good working environment and the possibility to get in touch with state-of-the art robotics research.
- prior experience with C++ is beneficial
- optimal control knowledge is beneficial but not required.
- prior experience with C++ is beneficial - optimal control knowledge is beneficial but not required.
Interested candidates can contact me via email: mgiftthaler (at) ethz (dot) ch.
Interested candidates can contact me via email: mgiftthaler (at) ethz (dot) ch.