How Do Autonomous Cars Deal with Double-parked Vehicles?

Home > News > Content

【Summary】For autonomous vehicles to become self-sufficient on public roads, they must be fully capable of performing complex driving maneuvers without human intervention.

Michael Cheng Aug 09, 2019 5:05 AM PT

How Do Autonomous Cars Deal with Double-parked Vehicles?

For autonomous vehicles to become self-sufficient on public roads, they must be fully capable of performing complex driving maneuvers without human intervention. Examples of such unpredictable scenarios include the following: merging lanes, getting around double-parked cars, parallel parking and three-point turns.

A maneuver that will likely be crossed off that list very soon is dealing with double-parked vehicles in congested cities. Although it may seem impossible for driverless vehicles to take on such feats, self-driving startup Cruise has developed a cutting-edge solution that addresses this issue. The company, which is heavily backed by General Motors, tested the feature on the busy streets of San Francisco.

Looking Out for Cues and Signs of Double Parking

The first step of the maneuver is identification. That is, the autonomous car should properly classify the vehicle as a double-parked unit. According to the startup, there are several cues that driverless sensors look out for on the road during real-time assessment. Some of these signs are: status of brake and hazard lights, distance to an intersection and ability to see around the obstacle.

Behind the scenes, data gathered by the sensors are processed using robust computer-vision protocols. Machine learning is applied in deciding if the vehicle is really double-parked. Specifically, the startup leverages a recurrent neural network (RNN) to reinforce its autonomous platform.

After identification and assessment, the driverless vehicle executes the maneuver based on instructions. This step is carefully carried out, factoring in the unit's surroundings and limitations of the car. The driving maneuver is optimized using Model Predictive Control (MPC). Cruise use MPC algorithms for motion planning.

"Transformation is an incredibly dynamic process and as I told my team, if we get it right we will never be finished," explained Mary Barra, CEO of General Motors.

"We will continue to act with speed, with discipline, and integrity to drive the business performance we need to win in today's market and in the future."

Encounter Rates

In large cities, double-parked cars are a common sight. More specifically, in San Francisco (where the startup performed its tests) the rate of encounter is 24:1. This rate is much higher than other locations in the US, including Phoenix, Arizona (where Waymo is testing its self-driving fleet). With this in mind, the way Waymo treats double-parked scenarios is very different, compared to Cruise. When one of the company's driverless minivans detects a double-parked car, it tags it as a stalled vehicle. After identification, it simply looks for the best way to drive around it.

"In San Francisco, each car encounters construction, cyclists, pedestrians, and emergency vehicles up to 46 times more frequently than in suburban environments, and each car learns how to maneuver around these aspects of the city every day," said Rachel Zucker and Shiva Ghose, Software Engineers at Cruise.

There's no doubt that testing double-parked obstacles in San Francisco is more difficult than suburban areas. This could give Cruise an edge over its competitors, as busy cities are expected to accommodate autonomous vehicles before other locations.