Tesla's Autopilot system has come under scrutiny recently, after a driver using the system was killed when his Model S hit a white tractor-trailer on a Florida highway. More details have emerged from the accident, with investigators revealing the car was traveling 9 mph over the posted limit at the time of impact.
The accident occurred on May 7th, when a tractor-trailer pulled across US Highway 27A in Florida. According to a report from the National Transportation Safety Board, performance data from the Model S says it was traveling at 74 mph (119 km/h) at the point of impact, 9 mph (14 km/h) above the 65 mph (105 km/h) posted limit.
In a release put out by Tesla, the company said the combination of a bright sky and white trailer might have made the truck difficult to see, potentially contributing to the accident.
After slamming into the side of the truck, the car rolled around 300 ft (91 m) before hitting a telephone pole. Having broken the pole, it then carried on another 50 ft (15 m) before coming to rest in someone's front yard.
In a statement released in June, Tesla was at pains to mention the fact Autopilot is still a public beta. Since then Elon Musk has clarified, using his Master Plan: Part Deux to reassure the public every iteration of the system is thoroughly tested in-house, but the beta label is used to minimize complacency among drivers.
The investigation into Joshua Brown's death is ongoing, with Tesla cooperating with the NHTSA and NTSB to identify who was at fault.
Source: National Transportation Safety Board
It does highlight that autonomous driving will have a lot of scenarios that have to be accounted for and will need to coordinate input from multiple sensor types. Companies are throwing hundreds of millions into R&D of these systems and the best system on the market just missed seeing the broad side of a tractor trailer.
These systems when fully complete will be massive difficult to maintain codebases that will require a lot of onboard computing power and layers upon layers of expensive redundancy.
As they become reliable enough to /seem/ predictable people are going to let their guards down and be less alert like typical passengers so we will continue seeing cases like this for a while during that long transition.
The four components of autonomous cars are 1) The manufacturer. 'Nuff said, but most manufacturers are on top of their game even though 5-10 years behind Tesla. 2) The driver. Drivers are going to have to adjust to a new paradigm where they have to learn and follow a new set of rules. This is going to be tough. "Monitor on Psychology" did a theme issue on this early in 2015. This included ethical issues. 3) Car-To-Car communications. Cars need to talk to one another because this allows sharing of sensor information (giving an almost total situational view) and to allow cars to signal their intentions. Though no one talks about it, this one is incredibly important. 4) Markings. There is a need to improve highway and vehicle markings in such a way that there are more easily and accurately seen. No more faded stripes or even "featureless" semi trailers. Even the most minimal marking would have prevented the collision in this particular incident.
There is no rationale that supports holding back on the technology until it is "perfect." It will never be perfect and at even Tesla's simplistic level, it is already safer than without driver assistance. This is just a variation on the nitrates and nitrites in processed meat paradox. The added chemicals are well-known to cause cancer. But the number of cancer deaths is less than 1/4000th of the deaths from spoiled meat. There is an old engineering saw that "perfect is the enemy of excellent."