AI could make autonomous vehicles safer

Obstruction sensing is key to self-driving vehicles, but safety will be further improved by predicting pedestrian behaviour, according to Maya Pindeus of Humanising Autonomy. 

Predicting human and animal behaviour is not always easy for people, let alone technology.

There are innumerable variables at work. Physicists would long ago have been coining it at the races if they did not need to simplify their classical predictive model by eliminating so many variables the horses became spherical.

Deep learning

Humans have a lifetime to learn from other’s behaviour. With the considerable advances in processing power, computers can now also learn in an empirical way. With big data they can also learn by example – so called 'deep learning'.

Newer vehicles are already incorporating this, but with autonomous vehicles just around the corner, they would need to interact with pedestrians who even themselves may not know their mind.

Self-driving in context

‘People in different places behave very differently, such as London versus Vienna or Shanghai’, says Maya Pindeus, CEO of Humanising Autonomy (HA).  

Fusing deep learning AI models with psychological and behavioural models helps to predict people’s behaviour on the street, Pindeus explained to IG’s Jeremy Naylor. Add to this data on the context and culture of where the car is driving hones the accuracy of the behavioural predictions.

HA is a London-based startup which has built a context specific tool for predicting human intentions. As things stand, self-driving vehicles have difficulty determining what pedestrians are planning to do. Are they planning to cross the road? Run or walk, straight or at an angle? Are they looking at their phone and don’t even see the car? This is a particular problem in the dense environments of European cities.

Tesla and Google driving interest

Elon Musk has promised a completely automated Tesla by the end of this year. Tesla aims to use cameras to cope with object detection including pedestrians – reactive rather than predictive. Google has driven the most autonomous testing miles, but where, Pindeus asks. Is it in remote parts of Silicon Valley or is it in cities, where pedestrian behaviour is more of an issue?

The HA software can be built into the software stack of autonomous vehicles so a car can predict what a person might do on the street. The company has been working with Daimler on developing the software for autonomous and semi-autonomous vehicles.

HA is using the software-as-a-service (SaaS) business model. The software will be licensed on a per vehicle basis.  

Denne informasjonen er utarbeidet av IG, forretningsnavnet til IG Markets Limited. I tillegg til disclaimeren nedenfor, inneholder ikke denne siden oversikt over kurser, eller tilbud om, eller oppfordring til, en transaksjon i noe finansielt instrument. IG påtar seg intet ansvar for handlinger basert på disse kommentarene og for eventuelle konsekvenser som et resultat av dette. Ingen garanti gis for nøyaktigheten eller fullstendigheten av denne informasjonen. Personer som handler ut i fra denne informasjonen gjør det på egen risiko. Forskning gitt her tar ikke hensyn til spesifikke investeringsmål, finansiell situasjon og behov som angår den enkelte person som mottar dette. Denne informasjonen er ikke utarbeidet i samsvar med regelverket for investeringsanalyser, så derfor er denne informasjonen ansett å være markedsføringsmateriale. Selv om vi ikke er hindret i å handle i forkant av våre anbefalinger, ønsker vi ikke å dra nytte av dem før de blir levert til våre kunder. Se fullstendig disclaimer og kvartalsvis oppsummering.

CFDer er komplekse instrumenter som innebærer stor risiko for raske tap på grunn av giring. 79 % av alle ikke-profesjonelle kunder taper penger på CFDer hos denne leverandøren.
Du burde tenke etter om du forstår hvordan CFDer fungerer og om du har råd til den høye risikoen for å tape penger.
CFDer er komplekse instrumenter som innebærer stor risiko for raske tap på grunn av giring.