Register now After registration you will be able to apply for this opportunity online.
This opportunity is not published. No applications will be accepted.
Continuous-time Online Visual-Inertial Odometry for Fast Flights
CT-VIO for agile flights
Keywords: Robotics, Computer Vision
The continuous time (CT) trajectory representation in visual inertial odometry (VIO) has the advantage of facilitating the fusion of the asynchronous, and potentially shifted, camera and IMU measurements.
This is beneficial in the case of a hardware synchronized sensor suite is not available.
CT-VIO introduces a prior encoding the smoothness of the trajectory to be estimated.
This prior can help the pose estimations of fast flying drones whose trajectories are expected to be smooth.
Temporal basis functions, e.g. B-splines, are the most common choice for CT-VIO / CT-SLAM.
Recent works have proposed algorithms to speed up the computation of the spline derivatives. Also, other efficient spline functions, like Hermite spline, exist.
In this project, we will start with studying different spline functions in terms of efficiency. We will select the best candidate and use it to develop an efficient CT-VIO algorithm which runs online on resource-constrained quadrotors.
The continuous time (CT) trajectory representation in visual inertial odometry (VIO) has the advantage of facilitating the fusion of the asynchronous, and potentially shifted, camera and IMU measurements. This is beneficial in the case of a hardware synchronized sensor suite is not available. CT-VIO introduces a prior encoding the smoothness of the trajectory to be estimated. This prior can help the pose estimations of fast flying drones whose trajectories are expected to be smooth. Temporal basis functions, e.g. B-splines, are the most common choice for CT-VIO / CT-SLAM. Recent works have proposed algorithms to speed up the computation of the spline derivatives. Also, other efficient spline functions, like Hermite spline, exist. In this project, we will start with studying different spline functions in terms of efficiency. We will select the best candidate and use it to develop an efficient CT-VIO algorithm which runs online on resource-constrained quadrotors.
Develop an efficient CT-VIO algorithm capable to run online on our quadrotors (target platform is the Nvidia Jetson TX2). Benchmark the proposed algorithm against existing state-of-the-art VIO algorithms.
We look for students with strong computer vision and programming background (C++ preferred).
This work involves a final demonstration of the proposed CT-VIO algorithm in a closed-loop controller to track fast drone trajectories.
Develop an efficient CT-VIO algorithm capable to run online on our quadrotors (target platform is the Nvidia Jetson TX2). Benchmark the proposed algorithm against existing state-of-the-art VIO algorithms. We look for students with strong computer vision and programming background (C++ preferred). This work involves a final demonstration of the proposed CT-VIO algorithm in a closed-loop controller to track fast drone trajectories.