Indy Autonomous Challenge — Silviu Roberto Popitanu

01

Overview

I worked with the TII EuroRacing team on the Indy Autonomous Challenge between May 2020 and November 2021. The team developed er.autopilot 1.0, a complete autonomy stack for the Dallara AV-21: a 390-horsepower, drive-by-wire race car equipped with GNSS, LiDAR, cameras, and radar. The system avoided static obstacles, performed active overtakes, and ran above 75 m/s (270 km/h); the team finished second and third in the competition’s first two main events.

The software was organized as ROS 2 nodes communicating through Eclipse Cyclone DDS. Perception, localization, motion forecasting, local planning, and control formed the driving pipeline, while supervision and failure detection provided a separate, redundant safety layer. I was responsible for the central Supervisor and Failure Detection module and contributed to the Mission Planner—the components that decided what the car was allowed to do and whether it was safe to continue.

I later co-authored the paper er.autopilot 1.0: The Full Autonomous Stack for Oval Racing at High Speeds, which documents the architecture, its track performance, and the lessons learned from racing at the Indianapolis and Las Vegas Motor Speedways.

LiDAR point cloud of a racetrack corner showing the road surface, track boundaries, and safety barriers. — LiDAR point cloud recorded at the racetrack, showing the geometry available to the perception stack.

02

My contribution

Developed the central Supervisor that coordinated start-up and emergency-stop decisions from system faults, Race Control, the Mission Planner, and the pit-crew joystick.
Built the Failure Detection module to validate sensor ranges, monitor critical powertrain signals, detect ROS 2 node timeouts, and supervise the base-station connection.
Contributed to the Mission Planner finite-state machine for initialization, pit exit, racing, and pit entry, including speed, overtaking, and opponent-distance rules.
Integrated and validated the safety and mission-management software on the full-scale Dallara AV-21 with the multidisciplinary TII EuroRacing team.

03

System architecture

er.autopilot 1.0 followed a Perceive–Plan–Act architecture. Perception and localization described the car and its surroundings; forecasting and planning selected a safe trajectory; control converted that trajectory into steering, throttle, brake, and gear commands. A stack-wide safety layer monitored the complete system.

The er.autopilot software architecture, with Supervisor, Failure Detection, and Mission Planner highlighted as my contributions alongside Perception, Localization, Motion Forecasting, Local Planner, Controller, and the Dallara AV-21. — Adapted from Figure 2 of the er.autopilot 1.0 paper. Green modules indicate my areas of contribution.

04

Technical details

Supervisor

The central coordinator combined automatic faults, Race Control commands, mission state, and manual pit-crew input into start-up and emergency-stop decisions.

Failure detection

Layered checks covered invalid or out-of-range sensor values, engine and energy-system limits, node health, message timeouts, DDS liveliness, and the radio link.

Mission planner

An SCXML-defined finite-state machine generated high-level references for pit entry and exit, warm-up, speed limits, overtaking permission, and following distance.

MicroSupervision

Local checks inside individual ROS 2 nodes provided redundant monitoring of vehicle state, command feedback, and dependent modules; controller nodes could stop the car directly.

05

Technologies

06

Media

High-speed autonomous driving with the real Dallara AV-21 on the Indianapolis Motor Speedway oval during the 2021 Indy Autonomous Challenge.