Register now After registration you will be able to apply for this opportunity online.
This opportunity is not published. No applications will be accepted.
Learning the Sampling Distribution for Nonholonomic Systems
We are looking to increase the sampling efficiency of the sampling based planners by learning the right distribution.
Keywords: sampling, planning, machine learning,
Sampling-based motion planning (SBMP) has emerged as a successful algorithmic paradigm for solving high-dimensional, complex, and dynamically-constrained motion planning problems. The performance of SBMP is tied to the placement of samples in the regions that might improve the solution; a result uniform sampling is only able to achieve through sheer exhaustion. In this project, we would like to investigate possibilities to bias the sampling distribution in order to speed up the convergence of the algorithm. Such biasing can be achieved using variational autoencoders. Preliminary results have been shown in [1], however no results for the non-holonomic systems have been shown. In this project we would like to focus on wheeled robots that cannot move arbitrarily in the SE(2) space. A successful outcome of the project would achieve a speed up in the planning time in cluttered environments compared to standard uniform sampling based planning.
Literature:
- [1] Arslan, Oktay, and Panagiotis Tsiotras. "Machine learning guided exploration for sampling-based motion planning algorithms." 2015 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). IEEE, 2015.
- [2] Ichter, Brian, James Harrison, and Marco Pavone. "Learning sampling distributions for robot motion planning." 2018 IEEE International Conference on Robotics and Automation (ICRA). IEEE, 2018.
Sampling-based motion planning (SBMP) has emerged as a successful algorithmic paradigm for solving high-dimensional, complex, and dynamically-constrained motion planning problems. The performance of SBMP is tied to the placement of samples in the regions that might improve the solution; a result uniform sampling is only able to achieve through sheer exhaustion. In this project, we would like to investigate possibilities to bias the sampling distribution in order to speed up the convergence of the algorithm. Such biasing can be achieved using variational autoencoders. Preliminary results have been shown in [1], however no results for the non-holonomic systems have been shown. In this project we would like to focus on wheeled robots that cannot move arbitrarily in the SE(2) space. A successful outcome of the project would achieve a speed up in the planning time in cluttered environments compared to standard uniform sampling based planning.
Literature:
- [1] Arslan, Oktay, and Panagiotis Tsiotras. "Machine learning guided exploration for sampling-based motion planning algorithms." 2015 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). IEEE, 2015.
- [2] Ichter, Brian, James Harrison, and Marco Pavone. "Learning sampling distributions for robot motion planning." 2018 IEEE International Conference on Robotics and Automation (ICRA). IEEE, 2018.
- Literature research on sampling based planning, variational autoencoders
- Setting up the training framework and implementing the training pipeline (generating the training data, possibly collecting some datasets, choosing the proper machine learning tools)
- Evaluation of the trained decoder + encoder
- Benchmark against other non-learning based planning algorithms
- Literature research on sampling based planning, variational autoencoders - Setting up the training framework and implementing the training pipeline (generating the training data, possibly collecting some datasets, choosing the proper machine learning tools) - Evaluation of the trained decoder + encoder - Benchmark against other non-learning based planning algorithms
- Familiar with machine learning tools
- Excellent programming skills (Python, C++)
- Knowledge of planning algorithms is a plus
- Knowledge of ROS is a plus
- Familiar with machine learning tools - Excellent programming skills (Python, C++) - Knowledge of planning algorithms is a plus - Knowledge of ROS is a plus
Edo Jelavic, jelavice@ethz.ch
Lorenz Wellhausen, lorenwel@ethz.ch
Please send your grade transcripts and CV.