Stanford Car Can Learn How to Handle Unknown Driving Conditions

Autonomous vehicle safety has been grabbing some headlines recently, with Nvidia trumpeting its Safety Force Field (SFF) safety platform, and then being called out by Intel’s MobileEye unit for ripping the idea off by copying elements of its Responsibility-Sensitive Safety (RSS) system. The longer version of the story is that both of those frameworks, and others developed by researchers around the world, depend on vehicle control systems that can dependably execute driving maneuvers as needed.
However, most autonomous vehicle testing gets done on and assumes good weather and road conditions. To operate at the limits of performance under harsh conditions like wet or icy roads means that a great deal of additional sensor information needs to be incorporated into the AI systems in autonomous vehicles.

Taking It to the Track
The team took Stanford’s Shelley and Niki autonomous vehicles out on the track to test their new learning network against both a skilled amateur driver and a physics-based model that had been programmed with over 200,000 vehicle trajectories. Basically, all three achieved similar results. That not only means that autonomous vehicles can be programmed to perform at human driver levels, but that they can learn to handle difficult driving conditions through experience.
Researchers caution that the system doesn’t perform well in conditions it hasn’t learned. That means gathering enough data from a variety of situations will be a challenge going forward. Gerdes sees the true potential in blending physics-based and neural network models to take advantage of each of their strengths.
Continue reading

Perseverance Rover Records the Sounds of Driving on Mars
This is the first Mars rover with a microphone, and it captured the eerie sounds of Perseverance driving over the regolith during one of NASA's tests.

Chia Demand Is Driving HDD Sales, Keeping Seagate’s Factories Full
Western Digital and Seagate are seeing sharply increased demand thanks to cryptocurrency, and they're both pretty happy about it.

Tesla Built a Supercomputer to Develop Camera-Only Self-Driving Tech
Tesla is talking about what it sees as the next leap in autonomous driving that could do away with lidar and radar, leaving self-driving cars to get around with regular optical cameras only.

Tesla Rolls Out $200 Monthly Subscription for ‘Full Self-Driving’
Some vague language on Tesla's part means that vehicles marketed as having full Autopilot capabilities might need an additional $1,500 hardware upgrade to use FSD.