Controls engineering or Artificial Intelligence
A conveyor is a remarkable feature in lots of industrial scenarios, e.g., factories, baggage handling at airports (Vanderlande’s use case). Though it is easier to detect from 2D image, 3D information is more important for many other tasks such as mapping, localization, and navigation. Thanks to the latest development of multi-layer LiDAR, we can obtain a dense 3D point cloud of the environment. Thus, the goal of this internship is to detect the conveyor from the 3D point cloud, which is rarely investigated. We will rely on the state-of-the-art deep learning method to achieve this goal.One problem that needs to be solved for deep learning is a huge amount of annotated data for training. Labeling it manually requires lots of human labor, especially for the 3D point cloud. So, a focus of this internship will be the automatic labeling of training data through sensor fusion and an environment model. The pipeline would be: (1) Automatic labeling of training data, (2) Using the generated training data to train the neural networks, (3) Improve the labeling quality or tune the neural network to improve the performance, i.e., accuracy of the conveyor detection.
This internship is part of the FAST project. This project researches new Frontiers in Autonomous Systems Technology. It is a collaboration between the Technical University of Eindhoven and several non-competing partners including Lely (Farming innovators), Rademaker (supplier of industrial bakery lines) and Vanderlande. We all need robots that can navigate autonomously in dynamic environments. The FAST project enables that by making the robots understand their environment and reason about their actions for example when interacting with humans.
The mission of the Innovate department is to explore new technologies that can be interesting for Vanderlande and its customers. Our robotics vision is to enable autonomous robotic solutions that add value to our warehousing, parcel & postal and airport customers. This is done with research, development and smart collaboration on relevant technologies. Our focus areas are the mechatronics, navigation, cognitive ability, perception, human interaction and system development of robotic solutions.
Tasks / Responsibilities
The internship will be about three months, the time planning is:
- Literature review, investigation, and data collection (2 weeks)
- Building an automatic labeling model and using the data for training and validation (6 weeks)
- Testing, tuning and debugging (2 weeks)
- Test it in the Vanderlande use case using real data from a robot (3 weeks)
- Experience with reinforcement learning and/or other artificial intelligence algorithms is preferred.
- Experience with mobile robot programming and modelling is also preferred.
- Knowledge of ROS and the Gazebo modelling package is required. Else, knowledge of a similar software package or tool that will allow importing a conveyor model and generating the simulated lidar data is sufficient too (e.g. Unity).
- In the case of ROS knowledge, C++ or Python experience is required, otherwise experience in the language of the other to be used simulation package is required.
Do you recognize yourself in this challenging profile? And are you looking for an internship/graduation assignment in an organization that has been elected as “Best Employer” for years in a row? Please fill out the application form and upload your resume and cover letter. For more information, contact us by e-mail: firstname.lastname@example.org.