Register now After registration you will be able to apply for this opportunity online.
This opportunity is not published. No applications will be accepted.
SLAM for Robotic State Estimation
SnookerBot is a snooker playing robot platform at IfA. This project builds upon it by adding a mobile platform as a base for the current system to move around the table. The student, then needs to explore the state-of-the-art SLAM algorithms to implement one on the testbed for state estimation.
Keywords: SLAM, control, state estimation, robotics, computer vision
SnookerBot is a robotic snooker playing platform at the Automatic Control Laboratory (https://vimeo.com/335260829). It consists of a full size snooker table, a robotic arm equipped with a linear motor and a cue and a camera system for state estimation. This project constitutes the next important step in the development of this platform by moving the robotic arm to a moving platform to circle around the snooker table.
The student will need to set up this new system, given the mobile platform and the cue equipped arm, by taking into consideration its required height, overall practicality and stability. For the stability of the cue, a small mechanical support needs to be developed and attached to the arm to form a closed kinematic chain. As with human players, it is expected to increase the overall stability of the platform and reduce currently present vibrations when taking shots. After this is successfully achieved, an accurate and robust mechanism of estimating the position of the snooker balls on the table must be developed. The current setup consists of two cameras; an overhead camera with a large field of view angle to capture the whole table and a stereo cue camera attached to the robotic arm. The student will research and implement novel SLAM techniques to estimate the pose of the cue without the use of the overhead camera. Currently, the ball detection and state estimation happens using ArUco markers and a convolutional neural network based algorithm. This gives an estimate with high enough accuracy and can potentially be used as ground truth for training purposes. A simple web based application needs to be developed in the end (built upon the existing one) to show the accuracy of the taken shots for debugging purposes.
SnookerBot is a robotic snooker playing platform at the Automatic Control Laboratory (https://vimeo.com/335260829). It consists of a full size snooker table, a robotic arm equipped with a linear motor and a cue and a camera system for state estimation. This project constitutes the next important step in the development of this platform by moving the robotic arm to a moving platform to circle around the snooker table.
The student will need to set up this new system, given the mobile platform and the cue equipped arm, by taking into consideration its required height, overall practicality and stability. For the stability of the cue, a small mechanical support needs to be developed and attached to the arm to form a closed kinematic chain. As with human players, it is expected to increase the overall stability of the platform and reduce currently present vibrations when taking shots. After this is successfully achieved, an accurate and robust mechanism of estimating the position of the snooker balls on the table must be developed. The current setup consists of two cameras; an overhead camera with a large field of view angle to capture the whole table and a stereo cue camera attached to the robotic arm. The student will research and implement novel SLAM techniques to estimate the pose of the cue without the use of the overhead camera. Currently, the ball detection and state estimation happens using ArUco markers and a convolutional neural network based algorithm. This gives an estimate with high enough accuracy and can potentially be used as ground truth for training purposes. A simple web based application needs to be developed in the end (built upon the existing one) to show the accuracy of the taken shots for debugging purposes.
1. Get familiar with the DeepGreen snooker table setup, specifically the current state estimation module.
2. Setup the Mobile Base Platform and verify that the current system still functions
3. Develop and deploy the mechanical support part to decrease arm vibrations. Validate its effect with tests.
4. Research SLAM literature for algorithms that can be implemented given the system’s constraints.
5. Develop and implement the chosen algorithm on the test-bed
6. Improve the existing visual tool to debug taken shots
7. Conduct tests with the new setup and compare with the old system
**Required Qualifications:**
• Good programming skills (Python/MATLAB/C++/C)
• Knowledge of basics in CV
• Experience with Mechatronics recommended
• Experience with SLAM recommended
1. Get familiar with the DeepGreen snooker table setup, specifically the current state estimation module. 2. Setup the Mobile Base Platform and verify that the current system still functions 3. Develop and deploy the mechanical support part to decrease arm vibrations. Validate its effect with tests. 4. Research SLAM literature for algorithms that can be implemented given the system’s constraints. 5. Develop and implement the chosen algorithm on the test-bed 6. Improve the existing visual tool to debug taken shots 7. Conduct tests with the new setup and compare with the old system
**Required Qualifications:**
• Good programming skills (Python/MATLAB/C++/C) • Knowledge of basics in CV • Experience with Mechatronics recommended • Experience with SLAM recommended
Please send us an email :
smenta@control.ee.ethz.ch, akarapetyan@control.ee.ethz.ch
Please send us an email : smenta@control.ee.ethz.ch, akarapetyan@control.ee.ethz.ch