Hasta la vista, baby!

The Terminator


BetterAI is a powerful perception AI for driverless cars in city traffic. It is both certifiable and safe.

A Better Perception AI

BetterAI has all the perception capabilities that systems like AutoPilot require. On top of that, it has the capability to drive with foresight, so your driverless car is up to the task to drive in city traffic. BetterAI can detect and correct the mistakes that every AI system occasionally makes. With BetterAI, there’s no need for a driver anymore.

Drive with Foresight

Analyzing driverless operation in city traffic under the safety standard ISO 26262, we’ve found that perception AI needs to provide a much deeper understanding of the situation: driving with foresight. Driving with foresight means to understand what other people do and will probably do next.

Let’s look at some examples.

Action Recognition

With the bicyclist on their own lane and nothing in the way, the driverless car can safely speed up.

The bicyclist looking over their shoulder is a weak indication that they want to change lanes, so the driverless car should not not accelerate anymore.

With the bicyclist signalling a turn, the driverless car should slow down, to maintain a safe clearance for the bicyclist to cross its lane.

Action Forecast

With the lane of the bicyclist blocked, the driverless car should anticipate that they will likely change lanes and slow down.

Risk Estimation

Driving with foresight means to understand the risk of a situation. What if your driverless car could estimate continuously in realtime, what safety engineers would otherwise just have assessed only once? In ISO 26262, the severity of a potential crash is an integral part of the risk assessment. The bicyclist besides the driverless is unprotected, and any mistake made right now would imply a high severity.

The fault tolerant time interval is another integral part of the risk assessment in ISO 26262. With more clearance between the bicyclist and the driverless car, this interval will increase.

In ISO 26262, the risk assessment concludes with the controllability of a potential crash. In a driverless car, there is no one who could control it. And the bicyclist can only control it when they see it coming.

Make No Mistakes

Analyzing AI technology under the safety standard ISO 26262, we know that, unlike humans, AI will interpret unknown situations in weird and unforeseeable ways. BetterAI will recognize unknown situations, unknown things, and others doing something unknown. In any of these situations, BetterAI will show your driverless car the safest move to gracefully adapt to the unknown.

Let’s look at some examples.

Edge Case Recognition

A perception AI that has never seen a skater before does not have a chance to identify them correctly. The AI might predict a person standing aside, which the driverless car would understand as safe to pass by.

Our perception AI can understand the fact that a situation is unknown and even point out the unknown object or unknown behaviour. With this information, the driverless car would understand that in this situation, is not safe to proceed.

So, what’s the way out of an unknown situation?

Adapting to the Unknown

Our perception AI can understand the fact that a situation is unknown, but if the unknown object or unknown behaviour is sufficiently similar to something known, our AI will know. Our AI knows bicyclists, and with the information, that the skater and their motion is similar to a bicyclist, the driverless car can proceed safely.