Lockered and Loaded
With its innovative Transporter, AV startup Udelv rethinks the entire concept of parcel delivery.
For most of its history, the autonomous-vehicle industry has focused on passenger transport, attempting to replace humans behind the wheel with a digital “driver.” The bulk of the industry’s engineering resources continues to be aimed at core driving functions, such as perception, controls and localization. But this nearly singular focus on robotic taxi drivers ignores countless opportunities to automate other types of vehicle operators, most notably the e-commerce delivery person who recently left a package at your front door.
The startup Udelv was founded in late 2016 to address that use case. Based in Burlingame, Calif., the company’s engineering work began in March 2017. Udelv knows its delivery trucks need to traverse highways and byways safely, but that’s a secondary function. Its raison d’etre is to automate the tasks related to loading goods into a Class 2 vehicle, efficiently routing it on a day-long schedule of stops, and having specific parcels seamlessly arrive at a determined location. That presents an entirely different set of engineering and design challenges.
Akshat Patel, Udelv’s CTO, relies on the company’s partnership with MobileEye, announced in April 2021, to enable its Transporter vehicle to complete core driving functions. That allows Udelv’s 35 employees to focus on automating and robotizing the numerous tasks around managing loads and deliveries.
“When you enter an unstructured environment to make a drop-off, there has to be a provision for the customer to interact with the vehicle to grab their stuff,” Patel noted. “We solve it through a collection of in-house technologies for cargo robotics, cloud robotics, teleoperations and fleet management intelligence.” The technology also makes it efficient for humans to remove packages from the vehicle, albeit manually.
Udelv CEO Daniel Laury explained the primary challenge. “When it comes to multi-stop, you cannot afford to mix the goods up among 100 packages, let alone have somebody steal somebody else’s,” Laury told SAE Media. “You have to invent something that has never been done before, a system of intelligent lockers. That’s what we created.”
Udelv conducted thousands of deliveries with its first-generation vehicle that had fixed lockers. It didn’t work because the fixed lockers often were too small or too big for the intended cargo. “If you’re carrying small pills for Walgreens, you need small compartments,” Laury explained. “But if you’re delivering a muffler for an auto-parts company, which is 90 inches long, you need a big space.”
After Udelv trials revealed the problem of not having suitable locker sizes for the range of diverse products that needed to be delivered, the team went back to the drawing board. The result is the Transporter, which was unveiled at CES 2022. While the press and crowds at CES focused once again on the technology stack — 13 cameras, nine lidars, six radars, and other sensors — the Transporter’s core innovation is its adaptive shelving space.
The system uses four sliding surfaces — horizontally and vertically on both sides — creating a right-sized cubby to carry everything from a 58-inch television to a 12-pack of beer. If every package is uniformly small, there’s room for 80 compartments. “The shelving behind the vehicle’s main doors is adjustable so that a goods organization can have dynamic space and the merchant can, on the fly, decide how they want to configure different spaces,” said Patel. “The aperture is dynamic and it’s software controlled.”
The mathematics of common multi-delivery starts with Amazon setting the benchmark of 200 to 250 stops a day via a single truck. FedEx drivers commonly make 100 stops a day. The control algorithm for the Transporter’s dynamic partitions — supported by a range of off-the-shelf position sensors, occupancy sensing, and shelf-presence sensing — is part of a full hardware-software stack designed for multi-stop delivery vehicles. Moreover, it’s linked to Udelv’s custom cloud services.
The ground truth regarding the vehicle and every one of its packages is reported back to Udelv servers, so managers and users know the state of the delivery. Goods are tracked the same way Uber riders watch their ride make its way on an app map. The opportunity for optimizations is significant. “We started with the assumption that the growth of e-commerce deliveries is so enormous that, within that constraint of 10,000-pound [Class 2 GVWR] vehicles, how can we maximize volume and the number of stops, because merchants will want more rather than less,”Laury said.
The Udelv vehicle-hardware-software regimen is designed to be automotive grade and rapidly scale. Laury explained: “When you talk to Walmart or Kroger, their first reaction is to say, ‘This is great, but how fast can you deploy? We’ve got 5,500 stores in the country. So, if this is a one-off, we’re not interested.’”
The Transporter’s vehicle-based locker system’s motors, sensors and other components are built to automotive standards. Everything goes through extensive durability testing to prevent a stiff jolt from a pothole from misaligning parts. The testing regimen includes lab tests for thermals, humidity, water and dust. Further validation includes prolonged driving.
Udelv didn’t have any direct precedents on which to rely in developing the Transporter. Patel pointed to conveyor belts at airports, automated vending machines and the Tesla Model X “falcon-wing” door. Patel’s career includes stints working on autonomous driving at Tesla, manufacturing automation projects at GM India and manufacturing automation and power electronics/motor controller projects at Caterpillar for its mining-truck product line. He also worked at Apple on its secretive automotive project.
A world-class teleoperations system is integral to the system. It values low latency over high resolution. Patel explained that loading and delivery locations — docks, parking lots, and driveways — can be lawless. “The most convenient spots for the customer to come out and pick up a package from the vehicle aren’t left to the robot, because it might expect a rendezvous 100 feet away.”
There is no consumer-facing touchscreen or audio system on the Transporter vehicle. Instead, all inter- actions are handled via the Udelv smartphone app. Teleoperators don’t need a high-definition movie-quality image, but they must be responsive, Laury asserted. “We have one of the best, if not the world's best, teleoperation system," he asserted,"which is entirely proprietary with one of the lowest latencies: about 105-millisecond glass- to-glass at 2,500 miles distance.”
It’s designed with dynamic bitrate management and in-house load balancing algorithms that interlink multiple cellular services. The remote drivers aren’t steering or generating throttle and brake commands but instead employing so-called decision primitives. “You just have to tell the vehicle to end up in a certain place, and it figures out the rest,” Patel said.
Custom ECU, algorithms
The same remote strategy is applied to the cargo space. If a customer can’t figure out the mobile app, the teleoperator can remotely open the cargo space for them. The system is modularized so that the pod (the 3,000-lb.(1361-kg) dynamic and connected intelligent locker) is interchangeable between Udelv skateboards. But unlike traditional flat electric-vehicle skateboards, the Udelv skateboard has tall pillars on both ends. The pillars measure about two meters (6.5 ft.) high. There is a traffic-light detection camera and telecommunication antennas on the top of each pillar.
“Companies usually show a flat skateboard about two feet off the ground and tell everybody this is going to be their autonomous skateboard,” Laury said. “But nobody thinks about where sensors are going. The longer the vehicle, the higher the sensors need to be because you need to see around. And if you have the long footprint of a Ford Transit like this one, you want to have your sensors placed high.”
The skateboard also can travel without the pod, allowing Udelv customers to carry items such as heavy equipment or stacks of covered pallets that can be exposed to the elements. If the 3,000-lb. pod is put aside, the core vehicle can expand its payload to 5,000 lb. (2268 kg). Future versions of the pod are expected to be lighter. Udelv developed a custom electronic control unit (ECU) to integrate cargo robotics, cloud robotics, smart delivery routing, telematics, teleoperations, remote connectivity, autonomous driving integration, and vehicle communication with controls into an efficient system. Udelv manages separate algorithms for simulating fleet-level intelligence for optimizing routing, including the best time and location for EV charging.
The company has a modest rollout schedule, starting with testing the Transporter this year before shifting to a pre-production sample in 2023. Initial production vehicles are slated for 2023. The output will expand to 25,000 vehicles per year by 2028, the company said, at which time 50,000 Udelv autonomous multi-stop trucks are expected to be on the road. The first markets will be Arizona and Florida, where legislation and weather are favorable.
CEO Laury notes the $700-billion aftermarket auto parts business will be one of the major industries to adopt the company’s delivery systems. If and when self-driving, multi-stop delivery vehicles are widely deployed by the likes of AutoZone, NAPA Auto Parts and O’Reilly, their brake pads, CV joints, mufflers and emissions parts will arrive to customers via completely autonomous electric delivery robots.