Most people have heard how autonomous driving will improve mobility and safety, but did you know about the costs saved by autonomous driving? In this blog we will describe why costs will be lowered with the implementation of Level 2 – Level 4 (L2-L4) autonomous driving features. The main two types of costs that will be reduced drastically are insurance and driver costs.
Testing Teraki’s “RoI” pre-processing for video.
Often underrated is that novel L2-L4 features will affect the ability of current insurances to cope with an insurance product of the car, as the driver’s insurance premium is often affected by the functioning and malfunctioning of these functions, Why should a driver’s insurance premium be increased if the car whose autonomous driving features were recently upgraded is malfunctioning? This problem is expected to persist as most cars will deploy various levels of L2-L4. 1
Underrated as well, is that a successful deployment of autonomous driving features requires humans to drive better and also according to well accepted traffic rules. Therefore enabling drivers to improve on their driving style in real-time while monitoring the learning rate of the car is also of key importance. A US-based startup Affectiva an Human perception AI software company detects complex emotional states of humans, pioneering the understanding of human decisions behind the wheel.
But let’s first have a quick glance at the five levels of autonomy:
Level 1: The classic example is cruise control where the engine maintains a set speed on long journeys.
Level 2: The car can handle steering and acceleration/deceleration simultaneously, but the driver needs to take control. Under critical circumstances automation is limited to very specific situation.
Level 3: At this level of automation, drivers can be more relaxed and can use their phones but need to be attentive as the car might still need a human intervention.
Level 4: One of the most advanced system that is currently available. It can handle almost all the situations but with some limitations it works only in certain geofenced areas/certain traffic conditions.
Level 5: Requires no human intervention, the car drives on its own.
Below we write about the impacts of Level 2 – Level 4 driving and we will end with some first observations on full Automated Driving, a.k.a. “Level 5” (L5).
L2-L4 Autonomous Driving and Insurance Costs Saving
The L2-L4 features are impacting a driver’s driving behaviour, the car and emergency situations to be handled. The driver assistance systems act as additional safety ‘watchdogs’ to prevent fatalities. Such safety features make the driver in a car with L2-L4 feature a much safer driver compared with a driver that drives a car without these L2-L4 features. Insurance companies would logically, and as a result, offer lower premiums to customers with these safety features.
Theoretically the insurance costs should decrease with autonomous driving. However we believe, as cars and particularly their bumpers get more complicated the insurance costs will rise in the short term due to increasing cost of repair. As semi-autonomous cars will be produced in higher volumes, the insurance costs would start to decline as economies of scale would push the repair costs down.
Deloitte, in its 2019 insurance outlook report said: “The rise of connectivity has generated a massive amount of real-time data and turned the insurer’s relationship with policyholders from static and transactional to dynamic and interactive.”
As Autonomous Driving will become safer and more data will be generated which will help in better underwriting the policies, insurance companies could then reduce costs of their car insurance policies. The same way as a novel L2-L4 functionality is able to be trained in real-time to improve its safety standard. Needless to say that along the learning process many false positives of the system will still occur and it is important that the failure is indeed not attributed to the driver but to the car solely. In this aspect a dynamic and interactive interaction with the driver, or enablement of edge processing capabilities is of key relevance.
Improve security features with Semi-Autonomous Driving (L2-L4)
Autonomous Car parking, Forward collision Braking, Autonomous complicated lane switching in case of an uncertain situation are current features in L2-L4. Although the car cannot drive itself but the security features can certainly help prevent accidents. To improve road safety these features need constant optimization. Therefore, to fully use these features, the OEM’s must constantly optimize them and this can be done by acquiring high volumes of detailed data from the car. Such big streams of sensor data will improve the quality of ADAS-features. Therefore, a constant and qualitative collection of telematics, cameras or lidar/radar data followed by a continuous update of the corresponding models can then be deployed in the car which in turn will lower insurance premiums. Most importantly, it enables the drivers to adapt their behavior accordingly, it is important to have the interaction deployed already at the edge and in real-time.
What’s in it for OEMs?
The L2-L4 functionalities need to be trained first with lots of data and then the OEMs need to continuously update and maintain the trained model. The model can be updated by analyzing the new data by receiving new data from the existing cars. As more data is fed into the model, the model is optimized further and will perform with much higher accuracy. It’s a continuous learning cycle. Teraki plays here a vital role in training the models and keeping them updated for optimal performance in production.
Teraki helps in these situations to both increase the accuracies of the models in real-time by a more tailored data collection, enabling a further 10x speed-up of the application within the car and more than x6 reduction in RAM requirements while maintaining CPU resources always within safety functional operation. The speed-up is adjacent to the one obtained by CPUs and AI chips 2.
The distribution of the computation has to be continuously adapted along the various CPU, GPU and AI Chip computing units, not only as a factory setting but dynamically while learning the novel L2-L4 features. More importantly, balancing the required bit-rates from the 10 cameras and up to 3 LIDARs deployed along the car is a key challenge affecting the bus capacities. Teraki’s Developer Center enables to prepare appropriate pre-processing models that can directly be integrated into the target ECU or TCU computing units via OTA 3 along proprietary machine learning models of OEMs/Insurances.
Enabling to collect specific data related to various components and sensors such as LIDARs deployed in the car, enabling OEMs and insurances alike to offer a warranty service. Moreover, the proper utilization of electric batteries of cars is a major cost driver in the insurance cost. Monitoring the data enables insurances to incentivize drivers to drive better and more importantly compensate for driver-linked degradation of these components.
The Ultimate Future with Level 5 Autonomous Driving
Now we take a glance into the future of Autonomous Driving where different claims on its time of arrival have been made over the past years. While there is still some debate on this, we start to see the potential of Autonomous Driving.
Lowering Regular Insurance Costs
According to Bloomberg, car owners will be paying 60% less premium by 2030 due to full implementation of Autonomous Driving i.e. L5. That kind of saving is a massive decrease in car ownership cost and is bound to change car insurance and fleet ownership.
Additional Cyber Risks will add to Costs
Although auto insurance premiums might decrease, the threat to such complex car system will still require highest standard of security. To insure for additional cyber threats, the insurance policies need to include car cyber security as a risk for AD cars. This new risk will then add new streams of revenue for insurance companies.
Certainly, the rise of adversarial attacks has been affecting the discussion on safety, as the manipulation of a machine learning is possible to result in fatal results. The ability to counter these attacks will have a key impact on the insurance costs.
Driver Costs
Furthermore, driver costs are $13 / hour. This assumption is based on the average driver cost in the US for an Uber driver. This cost can be eliminated completely when using a fully autonomous vehicle as robotaxis will be replacing human driver resulting in a drastic reduction of ride sharing costs that would impact the auto industry in multiple ways described below.
How could Robotaxi’s (L5) impact Costs?
Imagine a scenario where you never have to own a car, robotaxis will be everywhere and will be much more efficient than a normal taxi. Theoretically, the robotaxi can be utilized up to almost 100% therefore robotaxis will reduce car ownership purely due to its cost advantage. As robotaxis will be more efficient than a taxi, it will be able to make more rides in a given day and generate a higher revenue.
On average 95% 4 of time the cars are parked and are only eating up the precious city space. The view on this is if people switch to robotaxi’s, it would have a ripple effect in terms of traffic, air pollution levels, more open spaces in cities and fewer Road casualties.
Elon Musk claimed that by 2020 one could order fully autonomous taxi on demand. Tesla is aiming to rollout 1 million of such taxis by the end of 2020. The feasibility of such mass implementation of robotaxi’s is still up for debate but nevertheless we are nearing the future where robotaxis can become the norm after surpassing a period of coexistence of cars equipped with L2-L4 features along ultimately a fraction of L5 cars.
So, what would be the cost benefit by using a robotaxi? Robotaxis could cost less than $0.2/mile 5 which is comparatively very low to average ride onwership cost of $2-$3/mile. This massive cost difference can be mainly attributed to the elimination of the costs of the driver.
The introduction of robotaxis doesn’t mean people will stop owning cars. Fully autonomous cars can be committed to a fleet and can earn money for their owners by utilizing the idle time. Secondly, large groups of people have a strong individual desire of steering a wheel and controlling the speed while others perceive it as a daily burden and a reduction of productivity.
Which side do you stand?
Sources
1. Tencent/BMW
2. https://techcrunch.com/2019/07/19/powering-the-brains-of-tomorrows-intelligent-machines/
3. https://www.airbiquity.com/news/press-releases/airbiquity-partners-teraki-provide-over-air-software-and-data-management-edge-data-processin
4. https://fortune.com/2016/03/13/cars-parked-95-percent-of-time/
5. https://www.youtube.com/watch?v=YiWbdZ8ItRs&t=103s