Register now After registration you will be able to apply for this opportunity online.
This opportunity is not published. No applications will be accepted.
Space manipulator control and debris catching via Reinforcement learning
The goal of this project is to control a chaser satellite, consisting of a space manipulator mounted on a free-floating base, with Reinforcement Learning. The chaser should be able to grasp another free-floating target, such as a second satellite or a piece of space debris, and stabilize the stack.
Keywords: Space robotics, Reinforcement learning
Reinforcement learning for robot control has advanced rapidly over the last years, bringing advanced levels of agility and dexterity to machines. Traditionally, the movement of satellite-mounted manipulators is treated as disturbances to the system, which the on-board attitude controller has to compensate.
In the case of grasping an object of similar size but perhaps unknown mass properties, the attitude control system has to take care of stabilizing the coupled system. The task of jointly controlling arm and spacecraft, grasping a floating target, and controlling the complete stack presents an excellent opportunity for model-free control methods. Through simulation, several corner cases can be exploited which would increase the robustness and efficiency of the controller. In the scope of the project, you would have to set up the simulation and formulate the RL problem to achieve the individual tasks. Evaluating the performance of the controller in terms of robustness and efficiency for the spacecraft's resources is a secondary objective.
Reinforcement learning for robot control has advanced rapidly over the last years, bringing advanced levels of agility and dexterity to machines. Traditionally, the movement of satellite-mounted manipulators is treated as disturbances to the system, which the on-board attitude controller has to compensate. In the case of grasping an object of similar size but perhaps unknown mass properties, the attitude control system has to take care of stabilizing the coupled system. The task of jointly controlling arm and spacecraft, grasping a floating target, and controlling the complete stack presents an excellent opportunity for model-free control methods. Through simulation, several corner cases can be exploited which would increase the robustness and efficiency of the controller. In the scope of the project, you would have to set up the simulation and formulate the RL problem to achieve the individual tasks. Evaluating the performance of the controller in terms of robustness and efficiency for the spacecraft's resources is a secondary objective.
- Simulation and RL pipeline setup
- Task 1: Attitude control of chaser satellite
- Task 2: Grasping free floating target with chaser satellite
- Task 3: Stabilizing and attitude control of chaser and target
- Evaluation of results
- Simulation and RL pipeline setup - Task 1: Attitude control of chaser satellite - Task 2: Grasping free floating target with chaser satellite - Task 3: Stabilizing and attitude control of chaser and target - Evaluation of results
- High motivation for the topic and goal-oriented working attitude
- Experience with Reinforcement Learning for robotics
- Confident programming experience in C++
- Interest in Space Robotics
- High motivation for the topic and goal-oriented working attitude - Experience with Reinforcement Learning for robotics - Confident programming experience in C++ - Interest in Space Robotics
Please send your CV, Transcripts of Records, and a short motivational statement to Hendrik Kolvenbach (hendrikk@ethz.ch)
Please send your CV, Transcripts of Records, and a short motivational statement to Hendrik Kolvenbach (hendrikk@ethz.ch)