Lyft Has AV Mapped Out
The race to introduce the first commercially viable autonomous vehicle (AV) is on. A number of companies are jockeying to be the leader given the high stakes, particularly for Uber and Lyft, the incumbent leaders in the current ride-hailing market. Handicapping the winner in this race is not easy but early indications suggest that Lyft is ahead of the pack.
Major disagreement in the development of an AV is on the software side and not so much on the hardware side. The main difference between an AV and a regular car comes down to the components: sensors, a fast computer, and the decision-making software that drives the car.
Going deeper, the AV technology can be divided into two stages: Hardware (building the car) and software (training the neural network). On the hardware side, an AV has to be aware of its surroundings and its location. While there are no industry standards for hardware, the current AV technology uses either a standalone or a combination of radar, LIDAR, and camera technologies — each with its own advantages and disadvantages:
· Radars use radio waves to map the AV’s surroundings at very low computational costs, but are not very accurate;
· LIDAR’s use laser light to create a detailed 3D map up to 200 meters in all directions, but take longer processing time and are expensive;
· Cameras are extremely detailed, but use massive amounts of data and processing power.
Technological Challenges
Accordingly, the developers have to find the optimal balance between mapping and processing power. The technology is already available in the industry, so the big question remains why hasn’t AV become a reality? The answer lies in the second stage of the technology — software.
The software for AV already exists, but it has to be “trained.” By train, we mean the AV technology is based on deep learning algorithms that mimic the brain’s decision making. The algorithm makes statistical decisions based on previous experiences; thus, it has to be exposed to almost every single possible situation, this is called the training period. The AV’s computer has to process the inputs from the visual sensors and the decisions made by the software. To reduce computational power, AV has been integrated with 3D detailed maps of entire cities to reduce visual sensors to mostly focus on dynamic objects like other cars and people. The way dynamic objects interact is data required to train the algorithm; an extensive exposure to this data set will allow AV to become a reality and commercially viable.To train the algorithm, companies have to log a large number of miles in an AV. Waymo has the lead in this with over 10 million miles driven on public roads, compared to Uber’s 2 million miles. The main challenge to make the technology functional is to increase the number of miles driven and the interaction of dynamic objects on the road.
Lyft’s Edge in the Race to AV
While all the companies have their own twist on the technology, it is apparent that Lyft’s AV technology has an edge over competing technologies because Lyft is able to create detailed augmented reality 3D maps that can be programmable using regular smartphone cameras and integrated into the AV algorithms.
Although Lyft is far behind in the number of miles driven, its technology has the potential to set it apart from the competition. Lyft’s acquisition of Blue Vision Labs allows Lyft to create detailed HD maps using smartphone cameras. These maps use augmented reality to store data on speed limits, stop signs, bike lanes, and all kinds of information about the roads, but more importantly, the mapping technology also stores the way dynamic objects interact, reducing the training period for an AV. The technology’s data collection can be done from any smartphone on the windshield of any car, which increases the exposure of public roads to the autonomous software. At run time, the technology allows the AV to know its surroundings, and if the visual sensors perceive something different from the 3D map, the AV can interpret them as dynamic objects and increase its prediction capabilities.
Lyft’s ride-hailing platform could provide the required training and data collection to distinguish in leading the AV race. With over 2 billion miles driven on the platform last year alone the platform has great potential for collecting road information. If every Lyft driver had their phone’s camera collecting road data, the algorithm could be intensively exposed to public roads and its training period could be considerably decreased. This is an advantage that Lyft has over Waymo. Although Waymo has a larger number of miles driven in an AV, Lyft has the potential to train the algorithm at a faster pace with fewer miles driven in autonomous mode. This means that city permits and regulations would not affect Lyft since the data collection would be done from regular human-driven cars. The ride-hailing platform could facilitate the AV process of understanding possible road situations at almost no extra cost.
For now, these technologies will continue to develop in tandem, with developers assessing the advantages and disadvantages provided by each technological approach to autonomous cars.
While Waymo may have more miles driven, the advantages provided by Lyft’s approach to developing AV technology will likely allow the company to achieve faster advancement more efficiently and with fewer miles driven. In short, Lyft will produce more, for less.
