Full autonomous driving will only become a reality if developers can test their vehicles on the open-road. The 1000 Miglia Autonomous Drive (1000 MAD) is a competition in Italy that helps participants do just that.
The initiative challenges competitors to navigate their vehicles autonomously, for as long as possible, across approximately 1000 miles (2,200km) of public roads. During the test, the vehicles must drive through cities, historical towns, motorways and urban roads. Participants must ensure a high-level of safety at all times, especially when interacting with other road users and infrastructure.
OxTS and the 1000 MAD
For the third year running, OxTS worked with the AIDA (Artificial Intelligence Driving Autonomous) team at the Politecnico di Milano to help them understand the localisation performance of their car – a Maserati GranCabrio Folgore!
The AIDA team had two aims during the drive:
- To collect data during the whole route that would subsequently help them…
- Improve AI algorithms
- Refine control algorithms
- Digitally recreate the streets that the vehicle traversed
- Enhance the onboard experience for future passengers
- To drive autonomously across at least 300km of the route.
To help them collect such data, and drive without human intervention, the car contained an array of sensors, such as LiDAR, Camera, Radar and an OxTS AV200 GNSS/INS. The sensors helped the vehicle to determine its position and understand its surroundings during the drive.
After the drive was complete, the Politecnico di Milano were kind enough to share the navigation data with us so we could analyse and share it.
Data Analysis
Key:
- Poor accuracy (uncertainty σ>0.8 m)
- Average accuracy (0.4m<σ<0.8 m)
- High accuracy (σ<0.4 m)
Without looking at the data too closely, we can see that approximately 42.5% is of high accuracy, 39.3% is average accuracy, and 18.2% is classified as poor. Therefore almost 82% of the data is sub 0.8m accuracy with almost half of the total dataset within the Universities high accuracy threshold of below 0.4 m accuracy.
Whilst there are still some improvements to be made, we would deduce that areas where we saw poor localisation accuracy is down to GNSS obstructions during periods where the car travelled through urban environments.
This data can be improved further a number of ways. By utilising a higher grade GNSS/INS like the RT3000 v4 or alternatively, combining the AV200 GNSS/INS data with some of our other technology like LiDAR Inertial Odometry (LIO). LIO uses distance information from a 360° field-of-view LiDAR sensor, to constrain position drift in urban canyons.
We are delighted to say that the University completed the drive successfully and are busy analysing the data they collected.
We would like to pass on our congratulations to all of the AIDA team at the Politecno di Milano.
If you would like to learn more about the OxTS GNSS/INS solutions mentioned in the blog contact us using the form below.
