Xconomy reports that Nutonomy, which is testing robocars in Boston, has run into a problem in that the rolling Marvins can't figure out what to do at stop lights when there's solar glare that makes it hard to see just which color the signal is.
Unlike human drivers, though, the cars with a silicon chip inside their heads have a backup: An actual, trained human driver who goes on all test runs and who can take over in such perplexing circumstances. As the company notes in its quarterly report to Boston transportation officials:
Our system is designed to infer that a traffic signal is displaying a red light until the sensors can confirm the presence of a green light. If the inability to confirm a green light could create a safety risk, safety drivers are instructed to take over manual control of the vehicle.
Also, the company is not just sitting on its cold, steely hands and ignoring the issue:
When nuTonomy encounters a real world technical challenge like solar glare interfering with our traffic light detection, we prioritize the development of permanent hardware and/or software solutions, and implement an interim operational solution. In this case, our developers have added low-light data collection in the training of our algorithms. We have also made hardware adjustments, such as changing the exposure of a camera or adding glare shields. In the meantime, we have trained our safety drivers to be aware of this issue and know when to take over manual control preemptively.