Register now After registration you will be able to apply for this opportunity online.
This opportunity is not published. No applications will be accepted.
Master/Semester Thesis: Autonomous MAV Exploration using Reinforcement Learning
The task of autonomously exploring an unknown environment is one of the most challenging tasks in robotics. In this thesis we would like to tackle this challenge using reinforcement learning.
More Information: https://bit.ly/2Ily1qG
The task of autonomously exploring an unknown environment is one of the most challenging tasks in robotics, as it requires all components, such as reliable and accurate pose estimation and safe and efficient path planning, to work in perfect unison without reliance on external positioning or instructions.
In this thesis, we will focus on developing a novel exploration planner. Traditionally exploration planners follow handcrafted rules that try to move an agent to positions where it can observe frontiers, the borders between known and unknown space, or to maximize an exploration gain. There is a large body of research in this topic and many different hand-crafted or derived exploration gain functions have been explored. However, many of these algorithms need to be tuned manually to fit a specific robot or environment, and often implement some predetermined exploration strategy. With the advent of deep learning we have now the opportunity to let the robot learn on its own what the best exploration strategy is. To that end we have prepared a simulation environment to train a robot in 2D and 3D exploration and are now looking for a motivated student to explore RL-based exploration.
The scope of the project is flexible and depending on whether this will be realized as a master or semester project, we will start in 2D, move to 3D and will try to deploy this on a turtle bot or MAV. We are looking for a student that is excited about reinforcement learning but also driven by the prospect of deploying this on an MAV or turtlebot and not scared working with C++ code (Simulator) and ROS.
More Information: https://bit.ly/2Ily1qG
The task of autonomously exploring an unknown environment is one of the most challenging tasks in robotics, as it requires all components, such as reliable and accurate pose estimation and safe and efficient path planning, to work in perfect unison without reliance on external positioning or instructions. In this thesis, we will focus on developing a novel exploration planner. Traditionally exploration planners follow handcrafted rules that try to move an agent to positions where it can observe frontiers, the borders between known and unknown space, or to maximize an exploration gain. There is a large body of research in this topic and many different hand-crafted or derived exploration gain functions have been explored. However, many of these algorithms need to be tuned manually to fit a specific robot or environment, and often implement some predetermined exploration strategy. With the advent of deep learning we have now the opportunity to let the robot learn on its own what the best exploration strategy is. To that end we have prepared a simulation environment to train a robot in 2D and 3D exploration and are now looking for a motivated student to explore RL-based exploration.
The scope of the project is flexible and depending on whether this will be realized as a master or semester project, we will start in 2D, move to 3D and will try to deploy this on a turtle bot or MAV. We are looking for a student that is excited about reinforcement learning but also driven by the prospect of deploying this on an MAV or turtlebot and not scared working with C++ code (Simulator) and ROS.
More Information: https://bit.ly/2Ily1qG
- Design and implementation of an RL system for environment exploration
- Deploy proof of concept on a robot.
- Design and implementation of an RL system for environment exploration - Deploy proof of concept on a robot.
- Motivated and independent
- Programming skills (Python, C++, ROS)
- Basic computer vision knowledge
- Basic deep learning knowledge
- Bonus: experience in planning or exploration
- Bonus: experience with existing deep learning and RL tools
- Motivated and independent - Programming skills (Python, C++, ROS) - Basic computer vision knowledge - Basic deep learning knowledge - Bonus: experience in planning or exploration - Bonus: experience with existing deep learning and RL tools
Send CV and transcript to:
- Marius Fehr - marius.fehr@ethz-asl.ch
- Alexander Millane - alexander.millane@ethz-asl.ch
Send CV and transcript to:
- Marius Fehr - marius.fehr@ethz-asl.ch - Alexander Millane - alexander.millane@ethz-asl.ch