Joe, Florian and Sebastian on the Indy Autonomous Challenge

Indy Autonomous Challenge Overview

• The Indy Autonomous Challenge is a $1.5 million prize challenge for autonomous racing, with the final race taking place on October 23rd at the Indy Motor Speedway. (2m5s)
• The challenge involved developing autonomous racing algorithms and testing them in simulation before deploying them on real race cars, which were modified for durability and could potentially reach speeds of 180-185 mph. (3m12s)

TUM Autonomous Motorsport Team

• The main challenge for the TUM Autonomous Motorsport team, who won the challenge, was the initial uncertainty about the race format, the car's capabilities, and the final sensor and compute setup, requiring them to develop adaptable software and test it extensively in a hardware-in-the-loop simulator. (5m29s)
• The TUM team plans to open-source the software they developed for the Indy Autonomous Challenge. (27m23s)

Indy Autonomous Challenge Cars

• The Indy Autonomous Challenge cars were assembled by Junos Racing and then outfitted with autonomous driving kits by a company called Stuff. (8m40s)
• The self-driving hardware and software used in the Indy Autonomous Challenge was not originally designed for racing, which led to some unforeseen challenges. (8m53s)

Software and Data Management

• The software used in the challenge needed to be capable of multi-vehicle racing, including defending a position and overtaking other vehicles. (11m11s)
• Teams in the Indy Autonomous Challenge were permitted to utilize any software they chose, including commercial, open source, or custom-built solutions. (14m40s)
• While most teams opted for ROS2 Foxy, TUM (Team Technische Universität München) gained a competitive edge by upgrading to ROS2 Galactic, which offered significant speed improvements, particularly with the ROS Bag 2 for data recording. (16m39s)
• To gather training and testing data, a significant challenge involved managing the high volume of data generated by high-bandwidth sensors, necessitating efficient data writing to disk. (17m49s)
• Each autonomous race car was equipped with three terabytes of high-speed NVMe SSDs to accommodate the large data storage and processing requirements. (18m19s)
• Approximately 20 gigabytes of raw data, primarily from LiDAR sensors, were recorded per 20-minute run, which was instrumental in training neural networks for object detection, particularly for other race cars. (18m53s)

Race Day Challenges

• To qualify for the Indy Autonomous Challenge race on October 23rd, teams competed in simulation races to demonstrate their capabilities. (12m10s)
• During testing two days before the final race event, the PoliMOVE team experienced a spinout in turn one. (23m37s)
• The TUM team increased their car's speed by one or two meters per second every half lap until the car spun out at approximately 220 km per hour. (24m36s)

Impact and Future of the Indy Autonomous Challenge

• The Indy Autonomous Challenge (IAC) has led to improvements in open-source software packages like ROS, Cyclone, Zeno, and Autoware, which are used in universities. (29m10s)
• A follow-up multi-vehicle race for the IAC will be held in Las Vegas in early January as part of CES. (30m1s)
• People interested in transitioning into robotics should start with small, open-source projects and consider building scale model race cars, with instructions available online. (30m51s)