Uber’s self-driving cars are a key to its path to profitability

KEY POINTS

  1. Uber’s self-driving car division, the Advanced Technologies Group (ATG), has taken a new approach to autonomous driving since a fatal crash involving one of its vehicles.

  2. Uber plans to launch its self-driving cars in pockets of cities where weather, demand and other conditions are most favorable.

  3. Ultimately, the new strategy is designed to help Uber drive down costs as it seeks to show investors it has a clear path to profitability.


Pilot models of the Uber self-driving car is displayed at the Uber Advanced Technologies Center on September 13, 2016 in Pittsburgh, Pennsylvania.


Uber has been struggling since its IPO last year.Its co-founder and ousted CEO Travis Kalanick sold all his stock and left the company’s board late last year. Uber sold its much fast-growing food-delivery Eats division in India and public markets have voted by sending the stock steadily down since its debut as investors question its path to profitability.

“The vision for growth is absolutely there. But growth where it makes sense.” Dara Khosroshahi, Uber’s CEO told CNBC’s Andrew Ross Sorkin during an interview at the World Economic Forum in Davos, Switzerland. Unsaid in Khosrowshahi’s statement is the pivot to profitability in a market environment that’s stopped giving loss making “unicorns” a free pass.


Khosrowshahi’s goal is to get Uber to profitability by 2021. And the sharpest arrow in the company’s arsenal to achieving profitability is also the least understood. Uber’s self-driving unit, the Advanced Technologies Group (ATG), has an estimated valuation of over $7 billion, representing more than10% of Uber’s current market cap of about $61 billion.


And yet, Uber’s management or even the analyst community rarely discuss it. But speaking to those in the know you get a sense that this group which houses Uber’s self-driving car ambitions is the real key to Uber owning the future of mobility, a space that’s now seeing fierce competition from tech and automakers alike.


So why the hub-hub around self-driving, especially for a money losing ride hailing platform like Uber?


Cost.

The driver represents the single largest expense in non-autonomous ride-sharing at 80% of the total per mile cost, according to estimates by research firm Frost & Sullivan. By removing the driver from the equation, fully autonomous vehicles dramatically lower the cost of a ride while boosting its addressable market. Already offering software as a service, Uber plans to take the bet further by making the cost of rides so low (between its fleet of human and robot cars) that vehicle ownership becomes obsolete.


If done right Uber’s looking at a sizable slice of a very big pie. Realistic estimates for autonomous ride-hailing are still tough given regulatory hurdles. Yet, investment firm ARK’s research suggests that the 10-year net present value of this opportunity exceeds $1 trillion today and should hit $5 trillion by 2024 and $9 trillion by 2029. Of note, ARK is historically bullish on next generation technology bets, using the thesis to make investments.


“ATG is a growth play for Uber,” said Eric Meyhofer, the head of ATG at Uber. He was among the 40 to 50 people, many from Carnegie Mellon’s Robotics Institute, who left academia to fulfill the promise of bringing to market the new concept of a robotaxi. In 2015, under then-CEO Kalanick, the company had an ambitious target of achieving autonomy at scale by 2020.


Learning from their much documented downfall thereafter, including a fatal crash involving a self-driving Uber car in Tempe, Arizona, Uber’s ATG has a new scaled down, multi-pronged approach for its self-driving ambitions.


A long-term vision with short term goals, Uber ATG is focused on limited geographic presence and limited “autonomous capabilities.” ATG doesn’t want to solve every self-driving problem, everywhere, all-of the time as they compete with the likes of Tesla, Alphabet’s Waymo, Lyft, GM and Didi Chuxing.


Instead ATG plans only to introduce self-driving to new markets when it’s technologically feasible, safe and cost effective.

“The goal is to create a cheaper, better and safer automated option for consumers using Uber’s ride-hailing service,” Meyhofer said, adding that the technology has to pass through three stages: developing, piloting and commercialization. Uber’s ATG unit is currently in the development stage.


Cheaper means reducing the cost per mile of a self-driven ride to below that of an UberX ride today. A goal that’s still “ways off,” according to Uber ATG’s spokeswoman.


Better indicates making the ride a luxurious experience while reducing customer wait times. To this end Uber, partnered with Volvo and Toyota to co-engineer what Meyhofer calls the most “opulent” self-driving ride experience on the market.


Goal three, and probably the most important for Uber’s ATG unit, is safety, which could make or break the company’s self-driving ambitions. Regulators could also cause delays over safety concerns.


Today, Uber tests self-driving cars on roads in Pittsburgh. But before a self-driving Uber car even hits the road, ATG performs multiple rounds of software simulations to make sure its nearly perfect.


“But this is a mistake,” Jeff Schneider, a former Uber ATG engineering manager, said. “The pull back in road testing across the industry is not what will make the technology better. Those who get back on the road and test will take the lead.”


Schneider left the company in 2018 to return to academia as a faculty professor with a specialization in machine learning and robotics at Carnegie Mellon’s Robotics Institute.


But Meyhofer said getting the software right is equally as important.


“We’ve amassed petabytes of data by now, probably far more than Netflix,” Meyhofer said.


Developing a self-driving vehicle has two components: the software (the driver) and the hardware (the vehicle to be driven), Meyhofer said.


Various ATG teams in Pittsburgh, San Francisco, Washington D.C. and Toronto are working on building 3D maps, databases for machines to learn from and creating software for “perfect driving.”


For testing, humans drive the Uber’s modified Volvo XC90 on the streets. The first pass on a route is to map out the area. The second pass by a human-driven vehicle through the previously mapped areas is for the “perfect drive,” or the best version of perfect human driving. The data collected from “good driving behavior” feeds into Uber’s self-driving algorithm to teach the software how to drive on its own in the mapped area.


In many ways building self-driving technology is like teaching a teenager how to drive, equipping them with all the essential information, rules of the road, driving temperament and hours of practice in the hope that there will be no incidents once a computer takes the wheel.


Engineers run through real-world simulations to test the software’s navigational capabilities in “auto” mode if left on its own. Each simulation leads to software tweaks to make the driving better.


Internally, the ATG units’ policies are scrutinized by their self-driving safety and responsibility advisory (SARA) board created in the aftermath of the Tempe crash. The board reviews, advises and suggests changes to ATG’s policies on a quarterly basis.


Finally, the self-driven vehicle is ready to take the road, but only with a human driver, known in the company as a “mission specialist” behind the wheel. While human driver provides no input unless required, they stay with their hands floating above the steering wheel, so engineers back at home base can compare simulation to actual performance.


“The ultimate north star for the company is Level 4 autonomy,” Meyhofer said.


The industry defines Level 4 as “attention off” driving, that is the vehicle can take control under most circumstances and performs all critical functions, even making decisions like when to change lanes, and using a turn signal, on its own. But a key point to Level 4 is that the vehicle cannot operate in 100% of the conditions and therefore a human is required.


Sounds ambitious? It is. But don’t expect Uber to be on the sidelines until it achieves complete Level 4 autonomy.


Meyhofer brought about a radical new approach to thinking about autonomy. Think about a self-driving car that’s completely autonomous, but only when making right turns on a predetermined route. In the Uber ATG-defined autonomous world, this is a “limited operational domain” vehicle that could be deployed as a “self-driven ride” in Uber’s fleet.


It also underscores the out-of-the-box thinking Uber has deployed to retain investor confidence in its future bets while it still struggles to make a profit in its current business verticals.


“I worry that ATG is throwing good money after bad. Even in my conversations with them there seems to be almost of a ‘we have to do it’ versus ‘we want to do it’ approach,” said Bernstein analyst Mark Shmulik. Shmulik rates the stock as “outperform” with a price target of $40, while estimating the company will grow at a 30% rate over the next three years.


“ATG is not trying to build a robot car,” Meyhofer said. The grand plan is to build a self-driving ride sharing service that’s better, cheaper and safer than available transportation options, and integrate it to complement Uber’s current human-driven fleet.


But competition is building fast around the company with Waymo, GM and several others all working on self-driving technology.


“Essentially it comes down to the way the market plays out,” Shmulik said. “If Waymo becomes the predominant Autonomous Vehicle Operating System, then Lyft (and their partnership with Waymo) should realize margin expansion faster than Uber.”


Uber’s strategy is to be selective about where it launches self-driving cars. Instead of launching everywhere, Uber plans to map pockets of various cities that fit the most favorable profile for a self-driven vehicle, taking factors like weather, population density and road conditions into account.


For example, the company has identified the residential neighborhood of Squirrel Hill in Pittsburgh for deployment of self-driving cars with plans to expand in future. There are similar pockets identified in San Francisco, Toronto and Dallas, while simulating simulations for each of those cities takes place first at ATG’s offices in Pittsburgh.


This is also where Uber outshines the likes of Tesla or GM’s Cruise unit, which don’t have comparable riding data to leverage. An opportunity for Uber, as it analyzes its own ride-hailing usage patterns to identify opportunities most conducive for the self-driving unit to offer a self-driven ride that’s cheaper than riding with a human.


This caps the big shift for the ATG in the last two years: a multi-pronged, cost-driven approach to self-driving where they aren’t trying to be all autonomous, everywhere, at all times. Instead, Uber plans to make bets where it’s the most efficient to deploy self-driving cars


The future of the Uber ride-hailing app is to offer a menu of services to get people and services around. That ranges from the human-driven UberX car to a self-driving car on a predetermined route to a fully autonomous vehicle that can take you anywhere.


Overall, the hope is that spectrum can help reduce Uber’s costs and bring it to the profitability investors have been looking for over the past year.


SOURCE:paper.li

Recent Posts

See All

Build simple fuzzer - part 1

First of all, we are learning here and this fuzzer is in no way going to be a proper tool used against real targets (at least not initially). This is why we are going to code it in python. For real to

Build simple fuzzer - part 2

In the previous part of this mini-series we’ve implemented a very simple fuzzer. As the main idea behind it is being an exercise therefore I don’t think it is capable of finding bugs in complex target