Elon Musk believes that his Tesla will have fully autonomous robotic vehicles by the end of 2020. He recently stated the following : “There are no fundamental problems left. There are many small problems. And there is the main problem - to solve all these minor problems and put the whole system together. "
Perhaps the technology that allows the car to travel without human intervention (what the industry calls " level 5 autonomy ") is developing quite rapidly. However, it is quite another matter to produce a car that can do this safely and legally.
Fundamental problems for safely bringing fully autonomous cars onto the road do exist, and we still have to solve them before we see robotic vehicles on the roads. Here are five of the more challenging obstacles.
1. Sensors
Robomobiles use a wide range of sensors to sense the surrounding reality, and to recognize objects such as pedestrians, other cars and road signs. Cameras help robomobiles see objects. Lidar uses lasers to measure the distance from objects to a car. Radars recognize objects and track their speed and direction.
All of these sensors transmit data to the vehicle's control systems or a computer, helping it decide where to steer and when to brake. A fully autonomous robotic vehicle needs a set of sensors that accurately detect objects, measure the distance to them, their speed, and so on, under any conditions and in any circumstances, without the need for human intervention.
Bad weather, heavy traffic, painted road signs all negatively affect the accuracy of the sensors. The radar used in Tesla cars is not affected by the weather as much, but there are still challenges to ensure that the sensors used in the fully autonomous vehicle can recognize all objects with the level of confidence required for safety.
For truly autonomous robotic vehicles to emerge, these sensors need to work in all weather conditions anywhere in the world, from Alaska to Zanzibar, and in busy cities like Cairo or Hanoi. Accidents with current Tesla cars ( only working on the second level of autopilot ), including collisions with parked cars in July 2020, show that a company needs to bridge a large chasm to realize such a global and all-weather opportunity.
2. Machine learning
Most autonomous machines will use artificial intelligence (AI) and machine learning (ML) to process data coming from sensors and to help make decisions about further actions. These algorithms will help to determine the objects detected by the sensors, classify them according to training - this is a pedestrian, this is a traffic light, etc. The car will then use this information to decide whether to take any action, such as braking or roll-back, to avoid a collision with the detected object.
In the future, cars will cope with this recognition and classification more efficiently than a human driver. But so far, there is no generally accepted basis that guarantees that the ML algorithms used in robotic vehicles are safe. There is little consensus within the industry and among regulators on how to organize, test, and approve IO.
3. Open road
Once on the road, the robotic vehicles will continue their training. They will drive on new roads, recognize objects that they did not meet in training, and update software.
How to ensure that the reliability of the system remains at the same level? We need to be able to demonstrate that any new learning is safe and that the system does not forget the previous safe behavior. The industry has yet to reach agreement on this issue.
4. Rules and regulations
There are no satisfactory standards and guidelines for autonomous systems in any industry. Current safety standards for existing vehicles require a driver to take over in an emergency.
In the case of robotic vehicles, operating rules are already beginning to appear concerning certain functions, for example, automatic lane keeping . There is also an international standard for autonomous systems, which includes robotic vehicles, which sets certain requirements, but does not solve the listed problems with sensors, MO and operational training - although, perhaps, in the future it will develop to such a state.
Without recognized rules and standards, no robo-car, no matter how safe it may be considered, will be able to enter public roads.
5. Public acceptance
Tesla's current automated vehicles , like other automated and autonomous vehicles , have already been involved in various high-profile accidents. Public acceptance is a question not only for those who want to buy a robotic car, but also for those who will share the road with them.
The public should participate in making decisions about the introduction of robotic vehicles. Without this, we risk getting rejected by technology.
The first three of these problems must be addressed in order to overcome the last two. Naturally, there is a race for first place among the companies that have introduced a fully autonomous vehicle. But without working together to ensure security, presenting evidence of this security, working with regulators and the public to ensure
receiving approval - without all this, robotic cars for many more years will drive only on polygons.
While this may sound frustrating to entrepreneurs like Musk, the road to solving robotic vehicles is through long-term collaboration on these complex issues with safety, guarantees, regulations and approvals in mind.