Register now After registration you will be able to apply for this opportunity online.
This opportunity is not published. No applications will be accepted.
Optimizing Quadrotor Trajectories for Robust Vision-based Flight
In this thesis, you will develop a method to generate trajectories that reach a desired target location while guaranteeing a good position estimate throughout the flight considering the available visual features in the surroundings of the robot.
Keywords: Trajectory Optimization, Dynamic Programming (DP), Rapidly-Exploring Random Tree (RRT), Reinforcement Learning (RL)
In robotic systems that use Visual Odometry for localization, the quality of the position estimate depends strongly on the position of the quadrotor. This is due to the fact that its surroundings define the visual features based on which the position estimate is computed.
Approaches like SLAM build maps from observed visual features which enable the evaluation of the position estimate quality for a particular drone position offline. In this thesis, you will leverage such maps to generate trajectories that reach a desired target location while guaranteeing a good position estimate throughout the flight. Your task is to develop and implement a cost function that optimizes for these objectives and solve it with an appropriate method (e.g., Dynamic Programming (DP), Rapidly-Exploring Random Tree (RRT), Reinforcement Learning (RL)).
The thesis is done in collaboration with Tinamu Labs, a young and dynamic startup with the focus of bringing robotic technology into the real world. You will have the opportunity to test your algorithm with Tinamu’s robotic system and collaborate with their engineers.
**Tasks**:
In particular, the student will:
- Review literature on trajectory optimization and path planning algorithms used in Robotics with a focus on trajectory optimization for quadrotors.
- Create a cost function and an optimization approach for the given task. We will explore approaches based on state space discretization (e.g., Dynamic Programming, RRT, RL).
- Verify the concept in a simulation environment.
- Test and evaluate the algorithm in the real-world.
**Requirements**:
We are looking for an independent student who:
- Is motivated to work with a real-world robotic system.
- Has basic knowledge in optimization or other planning methods.
- Has background knowledge in Systems Control.
- Has experience with Python or MATLAB (C++ is a plus).
In robotic systems that use Visual Odometry for localization, the quality of the position estimate depends strongly on the position of the quadrotor. This is due to the fact that its surroundings define the visual features based on which the position estimate is computed.
Approaches like SLAM build maps from observed visual features which enable the evaluation of the position estimate quality for a particular drone position offline. In this thesis, you will leverage such maps to generate trajectories that reach a desired target location while guaranteeing a good position estimate throughout the flight. Your task is to develop and implement a cost function that optimizes for these objectives and solve it with an appropriate method (e.g., Dynamic Programming (DP), Rapidly-Exploring Random Tree (RRT), Reinforcement Learning (RL)).
The thesis is done in collaboration with Tinamu Labs, a young and dynamic startup with the focus of bringing robotic technology into the real world. You will have the opportunity to test your algorithm with Tinamu’s robotic system and collaborate with their engineers.
**Tasks**:
In particular, the student will:
- Review literature on trajectory optimization and path planning algorithms used in Robotics with a focus on trajectory optimization for quadrotors.
- Create a cost function and an optimization approach for the given task. We will explore approaches based on state space discretization (e.g., Dynamic Programming, RRT, RL).
- Verify the concept in a simulation environment.
- Test and evaluate the algorithm in the real-world.
**Requirements**:
We are looking for an independent student who:
- Is motivated to work with a real-world robotic system. - Has basic knowledge in optimization or other planning methods. - Has background knowledge in Systems Control. - Has experience with Python or MATLAB (C++ is a plus).
Not specified
Does this topic sound interesting to you? Please contact us!
balula@control.ee.ethz.ch
dliaomc@control.ee.ethz.ch
christoph@tinamu-labs.com
stefan@tinamu-labs.com
Does this topic sound interesting to you? Please contact us!