Register now After registration you will be able to apply for this opportunity online.
This opportunity is not published. No applications will be accepted.
Multi-LiDAR Robot Localization and Mapping
This project aims to enable the usage of multiple LiDAR sensors by developing a scan-to-”representation” registration. This can pave the way for going away from sensor specific and single-sensor scan-to-scan matching approaches.
Keywords: robot localization, lidar, point clouds, ANYmal, HEAP, mapping
The project will focus on the data registration from multiple LiDAR sensors to enable robust motion estimation of robots. LiDARs have become increasingly popular for their use as a sensor of choice for motion estimation and mapping in robots. However, the sparse nature of LiDAR point clouds, the narrow field of view of LiDARs, and the obstruction by the robot body limit the reliability and performance of the localization and mapping approaches. Due to miniaturization in sensors size and reduction in costs, if payload allows, robots such as ANYmal and HEAP are now fitted with multiple LiDAR sensors which can essentially be used to overcome limitations of single LiDAR sensors without investing in high-cost large FoV sensors.
Motivated by this problem, this project will focus on the registration of data from multiple LiDAR sensors and deal with challenges such as i) Arbitrary FoV and LiDAR configuration data utilization, ii) Non-overlapping LiDAR point cloud registration and, iii) Consistently increasing but Asynchronous LiDAR data usage. We propose the development of a scan-to-” representation” registration step instead of the traditional scan-to-scan registration to overcome these challenges and provide a reliable robot pose estimate and create an accurate map of the environment.
The output of the thesis will be utilized for hardware deployments conducted using both ANYmal and large outdoor construction robots, such as HEAP.
References:
[1] Yokozuka, Masashi, et al. "LiTAMIN: LiDAR-based Tracking And Mapping by Stabilized ICP for Geometry Approximation with Normal Distributions." IEEE IROS, 2020.
[2] Jiao, Jianhao, et al. "Robust odometry and mapping for multi-lidar systems with online extrinsic calibration." IEEE Transactions on Robotics (2021).
The project will focus on the data registration from multiple LiDAR sensors to enable robust motion estimation of robots. LiDARs have become increasingly popular for their use as a sensor of choice for motion estimation and mapping in robots. However, the sparse nature of LiDAR point clouds, the narrow field of view of LiDARs, and the obstruction by the robot body limit the reliability and performance of the localization and mapping approaches. Due to miniaturization in sensors size and reduction in costs, if payload allows, robots such as ANYmal and HEAP are now fitted with multiple LiDAR sensors which can essentially be used to overcome limitations of single LiDAR sensors without investing in high-cost large FoV sensors.
Motivated by this problem, this project will focus on the registration of data from multiple LiDAR sensors and deal with challenges such as i) Arbitrary FoV and LiDAR configuration data utilization, ii) Non-overlapping LiDAR point cloud registration and, iii) Consistently increasing but Asynchronous LiDAR data usage. We propose the development of a scan-to-” representation” registration step instead of the traditional scan-to-scan registration to overcome these challenges and provide a reliable robot pose estimate and create an accurate map of the environment.
The output of the thesis will be utilized for hardware deployments conducted using both ANYmal and large outdoor construction robots, such as HEAP.
References:
[1] Yokozuka, Masashi, et al. "LiTAMIN: LiDAR-based Tracking And Mapping by Stabilized ICP for Geometry Approximation with Normal Distributions." IEEE IROS, 2020.
[2] Jiao, Jianhao, et al. "Robust odometry and mapping for multi-lidar systems with online extrinsic calibration." IEEE Transactions on Robotics (2021).
- Literature research
- Formulation of registration representation and devising cost functions
- Implementation and validation with existing datasets
- Hardware experiments and performance comparisons
- Literature research - Formulation of registration representation and devising cost functions - Implementation and validation with existing datasets - Hardware experiments and performance comparisons
- Experience in C++
- Familiarity with ROS and point cloud processing (PCL or Open3D)
- Understanding of pointcloud registration techniques and optimization algorithms
- Experience in Python is a plus
- Experience in C++ - Familiarity with ROS and point cloud processing (PCL or Open3D) - Understanding of pointcloud registration techniques and optimization algorithms - Experience in Python is a plus