Lyft Outlines the Planning Model of Its Self-Driving Cars
Wed, April 21, 2021

Lyft Outlines the Planning Model of Its Self-Driving Cars

Lyft's model is inspired by Maslow and Asimov / Photo Credit: Volodymyr Plysiuk (via Shutterstock)

 

Like other startups and tech giants such as Google and Uber, Lyft is developing cars to hopefully ferry passengers sans drivers, a task that appears easier said than done, wrote Kyle Wiggers of VentureBeat, a transformative tech news platform. Lyft’s Level 5 division team is devoted to AV research and development, outlining a few challenges its engineers have encountered to date, “while pulling back the curtains on solutions and general progress.” 

Imagine this: “A car driving down a freeway encounters another car cutting across multiple lanes to make an exit.” How do we prevent a collision? The first car should slow down, but how dramatic should it be? The company’s self-driving prototypes tap a so-called “human-inspired” planning approach to find this out. Initially, Lyft used a baseline AI model that did not take into consideration the velocity of obstacles. However, its newest model naturally learned from human driving to slow down for vehicles “performing high-velocity cut-ins.” 

Lyft’s Level 5 division team wrote, “Ultimately, this blended approach resulted in a more comfortable and natural ride that was tuned to human preferences.” They believe that using “a combination of rule-based systems, learning-based systems, and human driving data will result in an overall system-level solution that exhibits the best of all worlds.”  Level 5 engineers said they were inspired by Maslow’s hierarchy of needs and Asimov’s three laws of robotics in developing the AV decision-making system. Safety and legality are at the base of their pyramid model. The company’s vehicles validate that planned behaviors are safe “and in compliance with local laws before executing them.”

The model also considers perceived safety to help reduce passengers’ and other drivers’ perceptions of being unsafe, regardless of when actual safety isn’t at significant risk. The next tier is comfort, which refers to aspects of rides like g-forces that might trigger nausea. The top tier is route efficiency. The Level 5 division said that the AV planner must correctly and consistently choose the right behavior. “Right behavior” doesn’t have any objective definition, “but utilizing the AV hierarchy of needs and then applying a mix of systems to address these levels of needs helps disambiguate this complexity,” they concluded.