Demo Day June 2024
On June 18th, the Autonomous Driving Lab held a Demo Day, showcasing the latest advancements and projects from its members.
Allan Mitt shared an exciting presentation about his experience mapping Tartu city in the CARLA simulator using drone footage. The level of detail in the simulation is remarkable — so much so that you might even spot your own parked car in the virtual version of Tartu! This development has promising potential applications, such as enabling driving schools to practice complex intersections without leaving the classroom and possibly even serving as a platform for driver exams in the future.
Here is a snippet from Allan's demo. Can you recognise the location?
Mahir Gulzar presented his experiences with testing route scenarios in CARLA. He shared with the ADL members an exciting option to simulate different types of scenarios on a single route, in contrast to the previous method of testing each scenario separately. Different trigger points will automatically initiate various scenarios for the route. For future work, he mentioned the need for automating trigger points and creating more scenarios.Tambet Matiisen shared his experiments with "bag scenarios". Instead of adding static, limited obstacles in the Autoware Mini environment, he proposed including dynamic obstacles automatically from recordings of ROS bag files. Such a simulation is very cheap to run and does not require expensive GPUs, making it suitable for regular laptops. However, it has some limitations, such as running ahead or behind a real car. Hence, for future work, appropriate performance metrics should be developed.
Dmytro Zabolotnii shared his latest work in the field of pedestrian prediction. He discussed his experience collecting state-of-the-art machine learning methods, implementing them into the Autoware Mini framework, developing novel metrics for performance evaluation, and collecting results from a database of .bag files and real-life experiments.
Edgar Sepp shared with the audience his latest experiments with one of the bag scenarios introduced by Tambet Matiisen. In real-time, Edgar experimented with different parameters and highlighted the various effects these parameters have on the simulation. This experimentation sparked a lot of interest from the audience, indicating that more exciting findings are to come!
Syeda Zillay Nain Zukhraf shared her latest work in localization using point clouds. She emphasized that in environments where GPS signals are unavailable, it is crucial to develop a robust localization method using LIDAR point clouds. Syeda walked the audience through the methodology of her work, current progress, and future steps, among which is validation of the solution.
Finally, Arun Singh shared some interesting insights from the International Conference on Robotics and Automation (ICRA). Those insights sparked lively discussion which smoothly flowed into a BBQ party on the terrace!