Bringing the reality of driverless buses one step closer

How can autonomous buses revolutionise city transportation? Jake Holmes finds out.

As technology advances and labour markets shorten, autonomous driving solutions for public transportation are becoming inevitable. Eran Ofir, CEO of autonomous driving company Imagry, sat down with International Transport Manufacturer to explain his company’s existing self-driving technologies, as well as the future of autonomous driving and how regulators can best help introduce these new systems.

HOW DO THEY WORK?

External 3D mapping is a method of capturing locations to provide a digital representation of the world. Although this technology has been used in autonomous driving before, it is not used to drive Imagry’s autonomous buses. Instead, a network of cameras feed information to the main computer allowing it to create a 3D map. Each camera is tasked with detecting traffic lights, vehicles, pedestrians and any other road-related items.

“We do not use those external HD maps that are provided to vehicles, but instead those that are created on the go,” says Ofir. “The entire system is built out of neural networks and is AI end-to-end. We get the video feed from eight cameras into an array of neural networks where each neural network is responsible for object detection and classification for a different type of object. One is looking at traffic lights, one at pedestrians, and then we build in real time a three-dimensional map that represents everything in the surrounding of the vehicle up to 300m and 360°.”

Imagry’s autonomous buses do not use rule-based coding, which is when computers are given specific rules to react to situations, similar to a decision flowchart. Instead, a black box is fed with information from the neural network which learns by imitation how to drive. Constant improvements can be made through this process, as the computer does not require additional programming to improve its driving.

“Through this, the system learns how to drive better and better, the same way I’m sitting in a vehicle with my daughter instructing her what she should and shouldn’t do whilst driving,” Ofir continues. “The future is mapless. No one can map the world in a resolution of 5cm as it is very costly and it will never be updated enough.”

Unsupervised training gives systems mass amounts of data and lets them lean how to operate through trial and error simulations. This can be risky and can potentially lead to a vehicle developing dangerous driving habits. On the other hand, supervised training of the system involves a person grading the decisions the vehicle makes. Supervised training can ensure the vehicle learns best practice for driving and safe operation.

“Because we are not using external maps but instead undertaking real-time processing, our latency, which is the most important parameter of autonomous driving, is shorter and we can respond faster to everything that happens on the road,” Ofir explains. “A system that is receiving external orders regarding how to drive on the road is prone to cyber-attacks by definition. A system that is self-sufficient will respond by itself and will not be prone to any threats from the outside.”

The only external instructions provided to the vehicle are directions, such as from Google Maps, which tells the vehicle where it needs to go, but it does not tell the vehicle what it expects to see on the road. By not using HD maps, the vehicle can detect and adapt to things that will not be on the map, such as roadworks and parked cars.

Bus detours are possible, as the self-driving features can work in any environment. This can be achieved without needing to provide a route beforehand. When the vehicle decides vision of the road is limited and it cannot conduct classification of objects, the system will not engage and safely stop.

TESTING THE TECH

Imagry tested its AI driver in Frankfurt, Germany. The vehicle had been brought from America and had never been driven in Germany before yet operated on the roads without any issues or hinderance. This was possible possible due to the company’s mapless technology.

Conversion of previous buses to Imagry’s new model is possible, but the steering box must support drive-by-wire. The driving system must be able to receive commands via its computer for driving. “Apparently, most electric buses do not have electric power steering,” Ofir explains. “An electric bus that was converted from diesel will not have it, as it will have the old power steering. Only electric buses that were born electric will have electric power steering.” 

NEXT STEPS

Imagry’s technology brings the reality of driverless buses a step closer, but despite having this technology developed, rolling out may still take some time. This is because sourcing buses and passing regulations remain obstacles for the technology.

How legislation impacts innovation

Imagry assisted in the creation of Israel’s autonomous driving bill and is currently working with German, French and Dutch regulators to create similar legislation. For the autonomous driving system to be activated on the road, the New Car Assessment Program (NCAP) must be passed. Currently, Imagry is so far the only company to pass NCAP for buses.

Ofir says: “We are the first and only company in the world that has passed NCAP testing for buses, meaning in 90 different scenarios of a bus driving between 30-60km/h - where all kinds of things are jumping out in front of the bus, and you have to stop and respond immediately - you have to pass with 100.”

Once the test has been passed, the system can be activated on the road but still requires a safety driver and cannot have any passengers at this stage. This is until UN Regulation 155 around cyber security is passed. This regulation ensures the bus cannot be remotely hacked and controlled externally.

Upon passing UN regulation 115, the bus can now operate with passengers, allowing it to be used for public transport.