Autonomous Vehicle Design Challenges

By Lindsay Farlow | Dec 11, 2013

If 90% of vehicles on the roads in the US were autonomous, fuel consumption would be reduced by 724 million gallons, 4.2 million accidents could be prevented, and 21,700 lives could be saved, according to a recent report from the Eno Center for Transportation. With statistics like that it’s no wonder that autonomous vehicle R&D is such a hot field.

Nuvation is leading that charge. We joined forces with an autonomous vehicle research team from the University of Waterloo and we’re already years into adding autonomous capabilities to DiscoFish, our own vehicle design. Legislative issues aside, there are many obstacles to overcome before autonomous vehicles are mainstream. Here's what we’re currently working on.

Coordinated Platooning

Platooning is a traffic system where vehicles travel together in a synchronized and densely packed group, communicating with wireless technology. This cooperative system increases traffic capacity on the roads, reduces fuel consumption, and improves safety by minimizing driver error.

Autonomous_Platooning.jpg
 

The research team is working with a new algorithm design called Cooperative Adaptive Cruise Control (CACC). This design handles vehicle-to vehicle communication and sensor noise issues, and can be used along with driver assistance algorithms such as lane departure warnings or merge-in assistance. The researchers are integrating systems of sensors with the CACC algorithms to guarantee autonomous vehicle stability, robustness, and performance. In the future this could mean that cars are able to fly safely down the freeways, bumper-to-bumper.

Coordinated Emergency Braking

The next autonomous design challenge is ensuring that vehicles are able to brake safely in case of an emergency. Current emergency braking systems are a combination of Autonomous Emergency Braking Systems (AEBS), Autonomous Brake Assist, and Advanced Driver Warning.

The AEBS system avoids or mitigates frontal collisions by using sensors to monitor the relative speed and distance of the vehicle ahead. The brakes are autonomously applied if sensor data indicates that danger is ahead, giving vehicles the maximum time to decelerate and avoid the collision. In vehicles operated individually, AEBS alone would not be reliable or practical, but when used with vehicle-to-vehicle communication, emergency braking becomes far more effective.

Autonomous_Braking.jpg
 

Emergency braking could be applied in different stages: advanced warning, pre-crash, and imminent crash (using the Advanced Driver Warning, Autonomous Brake Assist, and AEBS systems respectively). Advanced algorithms are used to adjust emergency braking parameters to optimize driving capacity while ensuring safety.

Object Detection and Tracking

Object detection for autonomous vehicles can be both LIDAR-based (using laser reflections to detect if there is an object) and/or vision-based (to detect a stop sign or traffic light). LIDAR is surprisingly good at object detection and mapping.

Lidar-Satellite.png
LIDAR vs. Satellite map

Salience detection strategies identify characteristics that distinguish objects from the background, and then Classification approaches try to label and understand them.  Classification algorithms can use template matching quite reliably for known objects like road signs, but it’s significantly more complicated for unknown or unexpected objects. Detecting traffic lights is particularly difficult because lights are not consistently visible and light detection requires cameras with a huge dynamic range.

Detecting and tracking dynamic objects requires a combination of point cloud data and visual cues. This brings the additional challenge of processing speed, as there are large numbers of targets and a large range of speeds. Sensor and algorithm robustness is an issue as well, as ice, snow, sun, rain, and shade all have different effects on different sensors. Researchers use predictive modeling to try and account for erratic behaviors and to detect anomalies.

As you can see, autonomous vehicles will require not just advanced sensors and vehicle-to-vehicle communication interfaces but extremely complicated algorithms and modeling techniques. Research teams are continuously improving their analyses, and experts predict that autonomous vehicles will be mainstream by 2016, if not sooner. Check back for more autonomous vehicle design updates and progress and developments from our research team!