Register now After registration you will be able to apply for this opportunity online.
This opportunity is not published. No applications will be accepted.
Reinforcement learning-based Active Localization
The goal of this project is to implement and test a reinforcement leaning-based localization system to allow an autonomous robot to perform construction tasks with high precision (in mm range).
Keywords: Artificial Intelligence, Reinforcement Learning, Robot Localization, Robot perception, Deep Learning, Autonomous Robots, Construction Robotics, Active Sensing.
In the construction industry, performing tasks such as drilling, chiseling or fastening with extreme precision is a vital requirement. Positioning the end-effector with **high accuracy** remains an open problem, as existing solutions rely on the engineering of the construction site, limiting the generalization and the applicability of the methods.
The aim of this project is to develop an **active localization system** using **reinforcement learning** algorithms, that allows high accuracy localization of the robot by relying exclusively on the onboard sensor suite. The goals are the following:
- Implementation of a reinforcement learning-based active localization strategy.
- Evaluation of the algorithm on the real robot and comparison against others state-of-the-art state estimators.
- Use of the localization system to perform a real construction task.
In the construction industry, performing tasks such as drilling, chiseling or fastening with extreme precision is a vital requirement. Positioning the end-effector with **high accuracy** remains an open problem, as existing solutions rely on the engineering of the construction site, limiting the generalization and the applicability of the methods.
The aim of this project is to develop an **active localization system** using **reinforcement learning** algorithms, that allows high accuracy localization of the robot by relying exclusively on the onboard sensor suite. The goals are the following: - Implementation of a reinforcement learning-based active localization strategy. - Evaluation of the algorithm on the real robot and comparison against others state-of-the-art state estimators. - Use of the localization system to perform a real construction task.
- WP1: Familiarization with current state-of-the-art literature on the active localization and reinforcement learning.
- WP2: Implementation of a reinforcement learning-based active localization system.
- WP3: Evaluation of the new framework with respect to other already existing strategies.
- WP4: Deploy and test the algorithms on cutting-edge robotic system.
- WP1: Familiarization with current state-of-the-art literature on the active localization and reinforcement learning. - WP2: Implementation of a reinforcement learning-based active localization system. - WP3: Evaluation of the new framework with respect to other already existing strategies. - WP4: Deploy and test the algorithms on cutting-edge robotic system.
- Good understanding of algorithmic challenges;
- C++ and Python programming experience;
- Knowledge of a Deep Learning framework is recommended;
- Knowledge in two of the following areas: Deep Learning, machine learning, localization, sensor fusion, computer vision;
- Students from outside D-MAVT (particularly, D-INFK, D-ITET, D-PHYS, D-MATH) are also highly encouraged to apply.
- Good understanding of algorithmic challenges; - C++ and Python programming experience; - Knowledge of a Deep Learning framework is recommended; - Knowledge in two of the following areas: Deep Learning, machine learning, localization, sensor fusion, computer vision; - Students from outside D-MAVT (particularly, D-INFK, D-ITET, D-PHYS, D-MATH) are also highly encouraged to apply.
Interested Students please send CV, Bachelor and Master transcripts to Luca Bartolomei (lbartolomei@ethz.ch).**Do not apply on Sirop.**
Interested Students please send CV, Bachelor and Master transcripts to Luca Bartolomei (lbartolomei@ethz.ch).**Do not apply on Sirop.**