After all, how do autonomous vehicles really go?

Share on facebook
Share on linkedin
Share on whatsapp

Recently there has been a lot of talk about autonomous vehicles and how they will transform the world for years to come. The benefits that this can bring are innumerable, from safety, reduction in the number of accidents, practicality, energetic economy, reduction of congestion, pollution, among others. Betting that this will become reality in the near future, major auto companies have been expanding their investments in the area to reach higher levels of automation of their vehicles. 

Despite the hype, few companies have already achieved a higher level of autonomy. In the market for the general public, the most advanced automation features are steering assistants, such as autopilot with centering of the track, for example the Tesla Autopilot and the Cadillac SuperCruise, where the driver can get his hands off the wheel in but you should be aware to enter the control at any time if the system fails. Despite being a very interesting feature, this is far from a fully autonomous level 5 vehicle in SAE J3016 international standard. 

The SAE International standard J3016 defines 6 levels of automation (or ADL – Autonomous Driving Level), the first level (ADL 0) being the vehicle without any automation, passing through steering assistance (ADL 1 and 2), partial automations under conditions (ADL 3 and 4) until reaching the last level of automation (ADL 5), in which the vehicle would be completely autonomous without needing any human interaction. 

The most advanced companies in this area are Tesla, which has been testing improvements in Autopilot, such as track change, and Waymo, which is ahead of the race, with a stand-alone taxi service in San Francisco in the United States. And some auto companies have already announced for 2019 the launch of vehicles with ADL 3, features, such as congestion mode. But this is a “simple” task compared to the challenges of the next levels of automation, ADL 4 and 5, which involve many issues and challenges not only in technological but also political, legislative and consumer acceptance issues. 

Safety is one of the hotly debated items that raises the most concerns whether fully autonomous vehicles will be able to identify and react to adverse situations in the best possible way. In a recent survey by the US Automotive Association, approximately 73% of Americans’ people feel insecure or afraid of a fully automated vehicle. 

Technologies 

Currently, most of the more advanced automotive research (ADL 4 and 5) uses, in addition to GPS location data, a set of radars, LiDAR distance sensors (Light Detector And Ranging) and cameras that map the environment around the car. These sensors generate an immense amount of data at any moment, and must be processed in time to make an adjustment in the vehicle. 

In the joint processing of sensor data, Deep Learning Neural Networks, Reinforcement Learning and Behavioral Cloning algorithms are used to identify elements in the environment, such as fixed components (poles, signs, etc.). ) and furniture (people, other cars, etc.), taking into account their trajectories, to then make a decision that may involve breaching, accelerating, changing lanes, parking, etc. 

For those who like more technical details about machine learning by reinforcement has this article, Robot explorer with learning by reinforcement, written by the colleague of Venturus André Sakiyama who explains how this works using a simple example of a robot made by him. 

One of the great challenges to the robustness of learning these networks is that our world is very complex, with many variations in the climatic and environmental conditions, such as snow, rain, fog, tarmac and dirt roads. There are many other challenges in the area, from hardware, sensors and infrastructure, such as coverage and quality of the 4G network, road quality, more recharging points for electric vehicles and many others. 

Brazil in this scenario 

Despite what many people may think, Brazil has been conducting several surveys with autonomous cars for almost 10 years, with prototypes such as CARINA – Robotic Car of Autonomous Navigation developed and tested at USP of São Carlos. In addition to the latest IARA (Intelligent Autonomous Robotic Automobile) developed at the University of Espirito Santo, which made a 74Km route from Vitoria to Guarapari in 2017. 

In addition, Brazil ranks as one of the 25 countries with the highest potential to receive this technology according to the AVRI (Autonomous vehicles Readiness Index) published in 2018 and 2019 by KPMG International. The index measures people’s acceptance, legislation, policy, technology, innovation and infrastructure. 

Considering the current automobile scenario, we still have several years until we have completely autonomous vehicles in the market for the final consumer. While this is not the case, Venturus seeks to update these new technologies to be part of this transformation, and has some case studies using technologies such as OBD-II (On Board Diagnostics) for measurement and analysis of vehicle parameters, with projected Hardware by Venturus with GPS, Wi-Fi, Accelerometer, Bluetooth and 3G / 4G. 

 

Source: 

SAE International, Surface Vehicle Recommended Practice, 2018. 

SAE Standards News: J3016 automated-driving graphic update, 2019. 

AAA believes testing, experience and education will aid consumer acceptance, 2019. 

KPMG Autonomous Vehicles Readiness Index, 2019. 

Cadillac Super Cruise, 2019. 

Tesla Autopilot, 2019. 

André Sakiyama, Robot explorer with learning by reinforcement, 2019. 

Veja mais
Means of Payment
Evens Taian

PIX is running

Have you ever imagined making and receiving money transfers, payments and any similar inter-banking transactions instantly and at any time? Soon, this will be a
Leia mais »
Campinas / SP - Brasil

Estrada Giuseppina Vianelli di Napolli, nº 1.185
Condomínio GlobalTech Campinas
Polo II de Alta Tecnologia
CEP 13086-530 – Campinas – SP
+55 (19) 3755-8600

+55 (19) 3755-8600
contato@venturus.org.br

Bitnami