Register now After registration you will be able to apply for this opportunity online.
This opportunity is not published. No applications will be accepted.
High-level Planning for Real-world Interactions of a Mobile Manipulator
Currently, our robot is commanded through a state machine that has to be adapted for every new task it encounters. In this project, we would like to replace the state machine with a symbolic planner that outputs a sequence of actions given a goal and the current world state.
Keywords: Mobile manipulation
Robotics
High-level planning
Symbolic planning
Perception
We use our mobile manipulation system “RoyalPanda” to interact with the world by rearranging objects or moving mechanisms such as doors and drawers. Currently, a state machine is defined for every task the robot needs to execute. It defines the high-level steps (navigate, grasp, place, …) the robot needs to follow to achieve the task, as well as the conditions under which it can switch from one step to the next. While this is a convenient method to experiment with few and short tasks, having to adapt it for every new task is inconvenient.
An alternative approach using symbolic planning [1] allows specifying a description of the robot’s capabilities, the current state of the world, and a goal state, and produces a plan. If followed, this plan leads the robot to achieving the specified goal.
The goal of this project is to implement symbolic planning as high-level planner on our mobile manipulation platform and demonstrate its capabilities with real-world tasks. To achieve this, we will combine existing robot skills (controllers for specific tasks e.g., grasping) with state-of-the-art perception algorithms, integrating everything in a modular framework with a symbolic planner. For this, we will also draw inspiration from existing systems [2, 3].
If you are excited about robots improving our lives in the future as well as state of the art perception, planning and control, we would be happy to hear from you.
[1] Karpas, Magazzeni, „Automated Planning for Robotics“, Annual Review of Control, Robotics, and Autonomous Systems, Bd. 3, S. 1–23, 2019.
[2] Curtis, Fang, Kaelbling, Lozano-Pérez, Garrett, „Long-Horizon Manipulation of Unknown Objects via Task and Motion Planning with Estimated Affordances“, arXiv:2108.04145, 2021.
[3] M. Bajracharya, et al. "A mobile manipulation system for one-shot teaching of complex tasks in homes." 2020 ICRA.
We use our mobile manipulation system “RoyalPanda” to interact with the world by rearranging objects or moving mechanisms such as doors and drawers. Currently, a state machine is defined for every task the robot needs to execute. It defines the high-level steps (navigate, grasp, place, …) the robot needs to follow to achieve the task, as well as the conditions under which it can switch from one step to the next. While this is a convenient method to experiment with few and short tasks, having to adapt it for every new task is inconvenient.
An alternative approach using symbolic planning [1] allows specifying a description of the robot’s capabilities, the current state of the world, and a goal state, and produces a plan. If followed, this plan leads the robot to achieving the specified goal.
The goal of this project is to implement symbolic planning as high-level planner on our mobile manipulation platform and demonstrate its capabilities with real-world tasks. To achieve this, we will combine existing robot skills (controllers for specific tasks e.g., grasping) with state-of-the-art perception algorithms, integrating everything in a modular framework with a symbolic planner. For this, we will also draw inspiration from existing systems [2, 3].
If you are excited about robots improving our lives in the future as well as state of the art perception, planning and control, we would be happy to hear from you.
[1] Karpas, Magazzeni, „Automated Planning for Robotics“, Annual Review of Control, Robotics, and Autonomous Systems, Bd. 3, S. 1–23, 2019.
[2] Curtis, Fang, Kaelbling, Lozano-Pérez, Garrett, „Long-Horizon Manipulation of Unknown Objects via Task and Motion Planning with Estimated Affordances“, arXiv:2108.04145, 2021.
[3] M. Bajracharya, et al. "A mobile manipulation system for one-shot teaching of complex tasks in homes." 2020 ICRA.
- Define perception requirements, review literature to find appropriate methods for achieving them. Start by simplifying perception using motion capture.
- Wrap existing robot skills into standardized interfaces
- Integrate with symbolic planning algorithm
- Deploy and test on the real system
- Define perception requirements, review literature to find appropriate methods for achieving them. Start by simplifying perception using motion capture. - Wrap existing robot skills into standardized interfaces - Integrate with symbolic planning algorithm - Deploy and test on the real system
- Highly motivated and independent student
- Interest in perception, planning and control
- Good programming skills in Python and/or C++
- Experience with the following is a plus: ROS, Git
- Highly motivated and independent student - Interest in perception, planning and control - Good programming skills in Python and/or C++ - Experience with the following is a plus: ROS, Git
Julian Förster (julian.foerster@mavt.ethz.ch)
Jen Jen Chung (jenjen.chung@mavt.ethz.ch)