by Hal Hodson, New Scientist
A communication system for autonomous vehicles would eliminate blind corners by enabling robotic eyes to perceive any unseen obstacle or person. With CarSpeak, developed by researchers at the Massachusetts Institute of Technology, human drivers approaching blind corners would not need to slow and prepare for the unexpected.
CarSpeak is designed to provide an autonomous vehicle with a continuous three-dimensional (3D) view of the area based on information captured by other cars, or by a sensor fixed in place to help with the blind spot.
The 3D view would consist of a cloud of millions of points generated by onboard laser-mapping equipment, but the system would allow cars to request specific sections of the area they are unable to see themselves. Other cars on the road would forward the information to the requesting vehicle.
Sending all the information to every car on the road would quickly overwhelm any wireless network. “There are hundreds of cars on the highway, and getting sensory data from all of them would be huge data congestion,” says Dina Katabi, project leader for CarSpeak at MIT. “But most of them want to know about what’s going on at the next exit.”
So, instead, CarSpeak allows cars to request a view of specific sections of the environment that they are unable to “see” themselves, with other cars passing information back to the requesting car. The system works out which regions are facing the most demand and then assigns more bandwidth to those sections as needed, speeding the process up.
In a test in Singapore, golf buggies running CarSpeak navigated the test environment more than twice as fast as those using simple wireless to relay data, and were 14 times less likely to collide with an unseen obstacle. Swarun Kumar, a lead researcher on CarSpeak, says that the next step will be a larger-scale test with multiple full-sized cars.
DCL: Lots of research and development to be done here before it goes real. Obviously event processing is an enabling technology, and CarSpeak will be doing a lot of real time event processing, but the researchers don’t seem aware of CEP as yet.