Register now After registration you will be able to apply for this opportunity online.
This opportunity is not published. No applications will be accepted.
Robot Action Planning for Spreading and Leveling Fluid-like Materials
Despite remarkable progress in robotic manipulation in recent years, manipulation of viscous fluids or granular materials, such as concrete, asphalt, or paint, remains a largely unsolved problem.
The goal of this project is to investigate planning methods for basic fluid manipulation tasks such as spreading and leveling fluids on a surface.
While manipulation of rigid bodies has shown impressive results recently, manipulation of viscous fluids and granular materials is still a largely unsolved problem, due to the difficulty in efficiently modeling and predicting their behavior for online planning.
However, the ubiquity of fluid-like materials in household applications or in the construction-, agriculture- and food industry, reveals the necessity for robots to have fluid manipulation capabilities.
Recently proposed approaches range from hand-crafted policies [2], over visual servoing techniques [2, 3] to policies that are learned end-to-end in an RL setting, either in simulation [1] or real-world [3].
In this project, we will focus on basic fluid manipulation tasks, such as spreading or leveling fluid-like materials on a surface.
The work packages include defining a set of basic skills (scooping, dumping, pushing, ...) and their parameterization, defining a suitable state representation (heightmap, image, contour map,...), building a transition model to approximate the fluid behavior, and investigate planning approaches to achieve spreading and leveling tasks in simulation.
Depending on the simulation results, the effectiveness of the approach will be demonstrated in a simplified real-world spreading task.
References:
1. Learning to Manipulate Amorphous Materials, Y. Zhang et al., 2020
2. Model-free vision-based shaping of deformable plastic materials, A. Cherubini et al., 2020
3. Learning Robotic Manipulation of Granular Media, C. Schenck et al., 2017
4. Robotic embankment: Free-form autonomous formation in terrain with HEAP, Jud et al., 2021
While manipulation of rigid bodies has shown impressive results recently, manipulation of viscous fluids and granular materials is still a largely unsolved problem, due to the difficulty in efficiently modeling and predicting their behavior for online planning.
However, the ubiquity of fluid-like materials in household applications or in the construction-, agriculture- and food industry, reveals the necessity for robots to have fluid manipulation capabilities.
Recently proposed approaches range from hand-crafted policies [2], over visual servoing techniques [2, 3] to policies that are learned end-to-end in an RL setting, either in simulation [1] or real-world [3].
In this project, we will focus on basic fluid manipulation tasks, such as spreading or leveling fluid-like materials on a surface.
The work packages include defining a set of basic skills (scooping, dumping, pushing, ...) and their parameterization, defining a suitable state representation (heightmap, image, contour map,...), building a transition model to approximate the fluid behavior, and investigate planning approaches to achieve spreading and leveling tasks in simulation.
Depending on the simulation results, the effectiveness of the approach will be demonstrated in a simplified real-world spreading task.
References:
1. Learning to Manipulate Amorphous Materials, Y. Zhang et al., 2020 2. Model-free vision-based shaping of deformable plastic materials, A. Cherubini et al., 2020 3. Learning Robotic Manipulation of Granular Media, C. Schenck et al., 2017 4. Robotic embankment: Free-form autonomous formation in terrain with HEAP, Jud et al., 2021
- Review literature on manipulating fluid-like materials, approximate fluid models, and suitable planning methods (heuristics, dynamic programming, reinforcement learning, optimal transport,...)
- Problem Definition: Skillset (scooping, leveling, pushing,...), state representation, transition model
- Investigate planning strategies for spreading and leveling tasks
- Verify approach in a simplified simulation environment
- Review literature on manipulating fluid-like materials, approximate fluid models, and suitable planning methods (heuristics, dynamic programming, reinforcement learning, optimal transport,...) - Problem Definition: Skillset (scooping, leveling, pushing,...), state representation, transition model - Investigate planning strategies for spreading and leveling tasks - Verify approach in a simplified simulation environment
- Highly motivated and independently working student
- Strong programming skills in C++ and Python
- Knowledge of system modeling and planning algorithms is required
- Knowledge of basic fluid dynamics would be highly beneficial
- Highly motivated and independently working student - Strong programming skills in C++ and Python - Knowledge of system modeling and planning algorithms is required - Knowledge of basic fluid dynamics would be highly beneficial