Register now After registration you will be able to apply for this opportunity online.
This opportunity is not published. No applications will be accepted.
Interaction of a mobile manipulator with articulated objects
In this work, we would like to design a controller for a mobile manipulation system that allows it to operate articulated objects and thus greatly increases the range of applications the system can be used for.
Keywords: Mobile manipulation, control, model predictive control, model learning
Robots being deployed in environments designed for humans need to be able to operate articulated objects such as doors, sliding doors, drawers or lids. The specifics of all such objects the robot may encounter throughout its lifetime cannot all be known when designing the robot. Therefore, it must be able to adapt to various types of mechanisms.
An approach that could achieve this would be reinforcement learning [1]. However, making it generalize over different instances of doors and drawers is challenging. Instead, we envision leveraging the fact that the number of mechanisms encountered is rather low. For each type of mechanism (prismatic, rotational, planar, free), a parametric model could be formulated [2, 3]. Upon encountering a new door, the robot would grasp it and start executing an initial guess trajectory. Measurements would then be used to fit one of the parametric models, which would in turn help improve the appropriate opening trajectory.
In order to be flexible regarding the spatial extent of the mechanism to be operated, the opening trajectory should be carried out using a whole-body controller, thus also making use of the degrees of freedom provided by the mobile base.
In the end, such a system should be able to open a wide range of doors, drawers and containers with only one or two attempts.
If you are excited about robots improving our lives in the future as well as state of the art control, we would be happy to hear from you.
[1] Gu, Shixiang, et al. "Deep reinforcement learning for robotic manipulation with asynchronous off-policy updates." IEEE International Conference on Robotics and Automation. IEEE, 2017.
[2] Karayiannidis, Yiannis, et al. "Model-free robot manipulation of doors and drawers by means of fixed-grasps." IEEE International Conference on Robotics and Automation. IEEE, 2013.
[3] Arduengo, M., Torras, C. and Sentis, L. (2019) ‘Robust and Adaptive Door Operation with a Mobile Manipulator Robot’, arXiv Preprint, pp. 1–14.
Robots being deployed in environments designed for humans need to be able to operate articulated objects such as doors, sliding doors, drawers or lids. The specifics of all such objects the robot may encounter throughout its lifetime cannot all be known when designing the robot. Therefore, it must be able to adapt to various types of mechanisms. An approach that could achieve this would be reinforcement learning [1]. However, making it generalize over different instances of doors and drawers is challenging. Instead, we envision leveraging the fact that the number of mechanisms encountered is rather low. For each type of mechanism (prismatic, rotational, planar, free), a parametric model could be formulated [2, 3]. Upon encountering a new door, the robot would grasp it and start executing an initial guess trajectory. Measurements would then be used to fit one of the parametric models, which would in turn help improve the appropriate opening trajectory. In order to be flexible regarding the spatial extent of the mechanism to be operated, the opening trajectory should be carried out using a whole-body controller, thus also making use of the degrees of freedom provided by the mobile base. In the end, such a system should be able to open a wide range of doors, drawers and containers with only one or two attempts.
If you are excited about robots improving our lives in the future as well as state of the art control, we would be happy to hear from you.
[1] Gu, Shixiang, et al. "Deep reinforcement learning for robotic manipulation with asynchronous off-policy updates." IEEE International Conference on Robotics and Automation. IEEE, 2017. [2] Karayiannidis, Yiannis, et al. "Model-free robot manipulation of doors and drawers by means of fixed-grasps." IEEE International Conference on Robotics and Automation. IEEE, 2013. [3] Arduengo, M., Torras, C. and Sentis, L. (2019) ‘Robust and Adaptive Door Operation with a Mobile Manipulator Robot’, arXiv Preprint, pp. 1–14.
- Review literature in manipulating articulated objects and whole-body control
- Set up simulation environment
- Implement a whole-body controller that allows a robot to operate different objects
- Test the system in simulation
- Integrate the system in a real-world mobile manipulation system
- Review literature in manipulating articulated objects and whole-body control - Set up simulation environment - Implement a whole-body controller that allows a robot to operate different objects - Test the system in simulation - Integrate the system in a real-world mobile manipulation system
- Highly motivated and independent student
- Interest in control
- Good programming skills in Python and/or C++
- Experience with one of the following is a plus: model predictive control (MPC), ROS, Git
- Highly motivated and independent student - Interest in control - Good programming skills in Python and/or C++ - Experience with one of the following is a plus: model predictive control (MPC), ROS, Git
Giuseppe Rizzi (grizzi@mavt.ethz.ch)
Michel Breyer (michel.breyer@mavt.ethz.ch)
Julian Förster (julian.foerster@mavt.ethz.ch)
Giuseppe Rizzi (grizzi@mavt.ethz.ch) Michel Breyer (michel.breyer@mavt.ethz.ch) Julian Förster (julian.foerster@mavt.ethz.ch)