Register now After registration you will be able to apply for this opportunity online.
This opportunity is not published. No applications will be accepted.
Design of a Dynamic Omni-Directional Kick Engine for NAO Bipedal Robots in RoboCup
This project aims to design a reliable omni-directional kick engine for the NAO bipedal robot, with dynamic optimization of motion parameters, based on nonlinear optimal control and reinforcement learning, as part of the NomadZ RoboCup Team at ETH Zürich (https://robocup.ethz.ch).
The design and implementation of a dynamic kick engine that is both fast and effective is required for competitive humanoid robot competitions like RoboCup. Using a NAO bipedal robot as a testbed, we will develop a dynamic kick engine that can generate an optimized kick trajectory with an arbitrary direction, while also guaranteeing closed-loop stability.
We are looking for an outstanding Master Student with a strong expertise in Robotics (e.g. Kinematics, Dynamics, Motion Planning), and Control Theory (e.g. Optimal Control, Model Predictive Control). Furthermore, a solid knowledge of programming in Matlab and C++ is required. Finally, solid experience in developing Reinforcement Learning algorithms (e.g. Q-learning) is desirable.
This project investigates a challenging yet important problem, and promising results will be published. Therefore, this topic is only offered as a Master Thesis Project (MA). Moreover, this project can be continued by the student and us in very interesting directions, involving further development on the robots as well as active participation in the team and future robot tournaments.
The design and implementation of a dynamic kick engine that is both fast and effective is required for competitive humanoid robot competitions like RoboCup. Using a NAO bipedal robot as a testbed, we will develop a dynamic kick engine that can generate an optimized kick trajectory with an arbitrary direction, while also guaranteeing closed-loop stability.
We are looking for an outstanding Master Student with a strong expertise in Robotics (e.g. Kinematics, Dynamics, Motion Planning), and Control Theory (e.g. Optimal Control, Model Predictive Control). Furthermore, a solid knowledge of programming in Matlab and C++ is required. Finally, solid experience in developing Reinforcement Learning algorithms (e.g. Q-learning) is desirable.
This project investigates a challenging yet important problem, and promising results will be published. Therefore, this topic is only offered as a Master Thesis Project (MA). Moreover, this project can be continued by the student and us in very interesting directions, involving further development on the robots as well as active participation in the team and future robot tournaments.
In overall, this master thesis includes the following parts:
1) In depth understanding of the different modules in the current codebase of the team.
2) Literature review of state-of-art approaches on walk-kick engine design and dynamic optimization of motion primitives based on reinforcement learning.
3) Comparison of different robot simulators, and selection of an appropriate one for the needs of the kick engine design.
4) Design of an effective kick engine with successful results on the selected robot simulator.
5) Even though various kick engines have been proposed in the literature, there are only a few works on how motion parameters can be optimized to give an effective kick. Achieving an accurate and powerful kicking requires a dynamic optimization of the speed and motion parameters of the robot. We will derive a reinforcement learning approach (e.g. based on Q-learning) to adjust motion parameters. Using reinforcement learning, the robot would learn to pursue an optimal policy to correctly kick towards designated points.
6*) Integration of the proposed kick engine on the codebase of the team, as well as implementation on the real NAO bipedal robots.
(Note: Parts with * are not mandatory, although they will be investigated based on the remaining time for completion of the thesis.)
In overall, this master thesis includes the following parts:
1) In depth understanding of the different modules in the current codebase of the team.
2) Literature review of state-of-art approaches on walk-kick engine design and dynamic optimization of motion primitives based on reinforcement learning.
3) Comparison of different robot simulators, and selection of an appropriate one for the needs of the kick engine design.
4) Design of an effective kick engine with successful results on the selected robot simulator.
5) Even though various kick engines have been proposed in the literature, there are only a few works on how motion parameters can be optimized to give an effective kick. Achieving an accurate and powerful kicking requires a dynamic optimization of the speed and motion parameters of the robot. We will derive a reinforcement learning approach (e.g. based on Q-learning) to adjust motion parameters. Using reinforcement learning, the robot would learn to pursue an optimal policy to correctly kick towards designated points.
6*) Integration of the proposed kick engine on the codebase of the team, as well as implementation on the real NAO bipedal robots.
(Note: Parts with * are not mandatory, although they will be investigated based on the remaining time for completion of the thesis.)
Interested students should send an email to Alexandros Tanzanakis (atanzana@ethz.ch) with an updated CV and transcript of records (both BSc and MSc).
Interested students should send an email to Alexandros Tanzanakis (atanzana@ethz.ch) with an updated CV and transcript of records (both BSc and MSc).