October 10, 2017 Read the article here

Today we are excited to release our first batch of videos to the general public, which recorded some of our routine tests. The videos tape 2 advanced scenarios: (1) unprotected left turns – a hallmark of top-level autonomous driving companies; (2) navigating congested road – driving like a veteran. These videos show Pony.ai’s technological ability and potential as a first-tier autonomous driving company.
Read the Chinese news coverage of these videos here.

Unprotected Left-Turn – A Hallmark of Top-Level Autonomous Driving Companies

Key observations: perception, inference, prediction

There are 2 types of traffic signals in California that permit a left-turn: solid green, and green arrow. When a green arrow signal is lit, the vehicle can perform a protected left-turn without waiting for oncoming vehicles from the opposite direction. When a solid green signal is lit, the vehicle needs to yield to those from the opposite direction before proceeding with an unprotected left-turn. At busy intersections where the vehicle needs to perform an unprotected left-turn, failure to detect any other road users can lead to severe traffic accidents, whilst false positives can lead to unnecessary long waits and obstruction of traffic.
It’s widely agreed on in the autonomous driving industry that whether a system can perform unprotected left-turns smoothly at busy intersections without human intervention is a very important indicator of the system’s maturity. This is a very technologically challenging scenario, especially at wide intersections with many lanes in each direction. The vehicle performing unprotected left-turn needs to yield its right-of-way to oncoming vehicles, therefore it must detect such oncoming vehicles correctly, monitor their movements in real-time, estimate their speed accurately, and anticipate their next moves. The autonomous vehicle also needs to mimic the driving behavior of human drivers, so that human drivers in other vehicles can understand and anticipate its intent. Pony.ai vehicles perform very well in this aspect.

  • 0:00
    Pony.ai vehicle detects and recognizes traffic signal correctly at 150 meters (~100 feet) away from the intersection, and accurately captures the whole process when the traffic signal turns from green to yellow, then to red.
  • 0:04
    Pony.ai vehicle turns on its left-turn signal at around 30 meters (~100 feet) distance from the intersection to signal its left-turn intent. Then it enters the left-turn lane.
  • 0:08
    Pony.ai vehicle reaches the intersection. The traffic signal turns from red to solid green. Pony.ai vehicle detected oncoming vehicles in the opposite direction in both the left-turn lane and the going-forward lane. Pony.ai vehicle creeps slowly into the intersection, obtains a good observation position without blocking any traffic, and monitors the movements of other vehicles.
  • 0:12
    After waiting for 2 oncoming vehicles to pass, pony.ai vehicle detects no more oncoming vehicles from the opposite direction, it starts its left-turn. Notice that pony.ai vehicle steers very smoothly, without any jiggling back and forth.
  • 0:23
    Pony.ai vehicle reaches the next T junction and prepares for another unprotected left-turn. This time, pony.ai vehicle is viewing the traffic signal against sunlight. Pony.ai vehicles have been tested in various time periods across the day, so that they can detect and recognize traffic signals correctly in different weather conditions, 24X7.
  • 0:56
    From a first-person perspective, there is heavy traffic from left-hand side to right-hand side in front of the pony.ai vehicle, and there is a vehicle blocking the path going forward. It’s a very complex scenario. When the traffic signal turns solid green, although there is still a long line of vehicles from the perpendicular direction, pony.ai vehicle can still find a way to creep forward and obtain a better perspective, in a fashion that’s familiar and understandable to human drivers, whilst obeying all traffic laws.
  • 1:02
    Pony.ai detects and monitors oncoming vehicles accurately, and predicts their movements correctly. After yielding to 5 vehicles from the opposite direction, pony.ai vehicles detects the 6th vehicle waits for its path to get clear. Pony.ai vehicle anticipates that the vehicle needs to wait longer to complete the left turn, it considers all conditions and decides to proceed with the left-turn. It finishes its left-turn smoothly.

Navigating Congested Roads – Driving like a Veteran

Key observations: perception, inference, prediction

Right-turn on red and following traffic during rush hours are both difficult scenarios for autonomous driving systems. When performing right-turn on red, the vehicle has the lowest precedence, it needs to detect all surrounding vehicles and pedestrians correctly and anticipates their movements. It needs to complete the right-turn without obstructing any other vehicle or pedestrian with higher precedence. Pony.ai vehicles utilize multiple types of sensors so that they can monitor their surrounding environment without blind spots. During a traffic jam, not only there are more than usual number of vehicles to pay attention to, each vehicle also starts and stops more frequently, so it’s very important for the autonomous driving system to control the vehicle steadily and react to the actions of the preceding vehicle quickly. Pony.ai system performs well in this aspect, as shown in the following video.

Some highlights of the video:

  • 0:04
    Pony.ai vehicle approaches an intersection. Traffic signal turns red. There is one person crossing the street on a skateboard from the left-hand side. Pony.ai vehicle stops before the crosswalk, waits for the skateboarder to pass, before it proceeds with its right-turn (0:08).
  • 0:09
    Vehicles keep coming from left-hand side, pony.ai vehicle yields to them according to the law.
  • 0:12
    After all oncoming vehicles have passed, pony.ai vehicle starts its right-run. Although the video shows part of the surrounding environment, pony.ai vehicle is capable of monitoring full 360-degree surrounding environment all the time.
  • 0:19
    Pony.ai vehicle merges into the afternoon rush hour traffic. It is able to detect and track the preceding vehicle and all vehicles around, and keeps a safe distance at all times. When the preceding vehicle moves slowly, pony.ai vehicle follows at an appropriately low speed.
  • 1:27
    On green light, Pony.ai vehicle turns right onto a narrower street. Note that this is a narrow street with quite many vehicles on the opposite direction. Pony.ai completes the turn very smoothly in this case, without any unnecessary jiggling in steering.
  • 1:33
    A cargo truck enters the road from a parking lot and approaches pony.ai vehicle. After considering the truck’s position, speed and its right-of-way, pony.ai vehicle decides to go forward with steady speed.