Dot (DoorDash)

Dot is an autonomous delivery robot designed and built in-house by DoorDash Labs, the robotics and automation arm of DoorDash. It was shown to reporters at DoorDash's San Francisco headquarters at the company's Dash Forward event on 29 September 2025 and announced publicly the following day, alongside an order-routing system called the Autonomous Delivery Platform.^[1][2][5][6] Dot is a four-wheeled, road-capable electric vehicle that travels on roads, bike lanes, sidewalks and driveways at speeds up to 20 mph, and it is explicitly not a sidewalk robot: DoorDash draws the contrast itself, writing that "Traveling up to 20 mph, Dot can ship items faster and much further than sidewalk robots," and its dispatch platform treats Dot and third-party sidewalk robots as separate delivery modes.^[1] DoorDash describes Dot as "the first commercial autonomous robot built to travel seamlessly on bike lanes, roads, sidewalks, and driveways, and specifically designed for local delivery," a first-of-its-kind claim the company has made on its own authority and that no independent body has adjudicated.^[2]

Scroll sideways for more →

Why DoorDash built its own robot

DoorDash Labs is led by co-founder Stanley Tang. By its own account, "in 2018, we founded DoorDash Labs," though the unit was not introduced publicly until 18 November 2021.^[28] TechCrunch counted seven years of work leading up to Dot: DoorDash acquired the autonomous-driving startup Scotty Labs in 2019 and brought on the co-founders of the AV mapping startup Lvl 5, then hired Ashu Rege from Zoox in 2021 to run autonomy.^[4] Scotty Labs was a tele-operations company, which matters for how Dot is supervised today. Bloomberg reported that DoorDash also hired dozens of former staff from General Motors' Cruise unit after that self-driving business was shut down, placing them in product design, policy, partnerships, recruiting and remote support of the robots.^[5]

Tang's stated rationale is that neither dominant autonomous vehicle form factor fits food and convenience delivery: a robotaxi-sized vehicle cannot bring an order to a front door, sidewalk robots reach the door but are slow and short-range, and drones carry further but only light payloads.^[6] The engineering post frames this as the "last ten feet" problem. Unlike ride-hailing, where passengers walk to and from the vehicle, delivery customers expect goods at the door, so the vehicle must be "nimble enough to handle sidewalks and driveways, but fast enough to get ice cream to you before it melts while navigating street traffic."^[1] Put plainly, a robotaxi's job ends at the curb; a delivery robot's hardest problem starts there.

Patent lineage

DoorDash's robot work is documented in patents that predate Dot's unveiling by years. Utility patent US 11,372,426 B2, "Automated vehicle for autonomous last-mile deliveries," claims priority to a provisional filed 5 December 2018, was filed 5 December 2019 and granted 28 June 2022. It describes a courier vehicle for sidewalks, bike lanes and roadways with "a maximum width of 36 inches, a weight between approximately 200-300 pounds, and travel a maximum speed of 18 to 25 miles per hour," a compartment "sized to fit two to three twenty-two inch pizza boxes," and a sensor module on a support structure "three to five feet" above the ground.^[25] Two design patents for an "Automated delivery vehicle," USD931,756 and USD932,368, were filed 17 May 2019 and granted in September and October 2021.^[26][27] Harrison Shih, later head of DoorDash Air, is a named inventor on all three, and all are assigned to DoorDash.

The shipped robot sits partly outside that envelope. Dot matches the 36-inch width and the 18 to 25 mph band, but weighs 350 lb rather than 200 to 300, holds roughly double the patented compartment's two to three pizza boxes, and carries its sensors on top of and below the storage compartment rather than on a raised mast.^[1][5]

Design and specifications

Dot's round main compartment was inspired by DoorDash's D-shaped logo and, per Bloomberg, resembles "a large cheese wheel"; it opens from the front like a baby stroller.^[5] Two circular light displays act as cosmetic "eyeballs" that look left and right depending on where the robot is navigating, a human-robot interaction choice that, according to Rege as reported by Bloomberg, helps make the machine more acceptable within the community it serves.^[5] Dot also has an LED text strip, speakers and a microphone, and a reporter at the launch observed the eyes glancing toward nearby voices.^[4][7]

Scroll sideways for more →

Per-surface speed caps apply where cities impose them. Fremont's encroachment permit sets a maximum of 5 mph on sidewalks, 16 mph in bike lanes and 20 mph on neighborhood streets.^[8] DoorDash's own description of Dot's driving style to New Mexico legislators was that it "behaves like a cautious cyclist."^[16]

The autonomy stack

Everything below comes from DoorDash's engineering post of 30 September 2025 unless otherwise noted. None of it has been independently verified; DoorDash has published no technical paper, benchmark, architecture diagram or safety case.

Perception

Dot uses what DoorDash calls a vision-primary approach: 8 external cameras giving 360-degree coverage, plus 1 interior camera "to ensure delivery quality," supported by 4 inexpensive radar units.^[1] Three high-resolution lidar sensors are also fitted for situational awareness, but the company says these "are being replaced by inexpensive automotive-grade lidars resulting in a low-cost sensor stack to enable rapid commercial scale."^[1] That is a cost decision rather than a capability one, and it says something specific about the scaling strategy: DoorDash is betting that a computer vision system trained on enough data can carry the perception load expensive lidar carries today, leaving cheap automotive lidar as redundancy rather than the primary range sensor. For a vehicle meant to be built in large numbers, sensor cost bears directly on whether the fleet economics work. Dot's current lidar units are custom parts the company wants to swap for off-the-shelf equivalents.^[7] The interior camera has a second purpose beyond quality control: DoorDash says it helps ensure children do not climb into the cargo space.^[4]

Planning: learned prediction with a classical safety net

DoorDash describes a hybrid planner: "Our stack combines deep learning and search-based algorithms to find a safe, smooth path through a complex world. Deep learning allows us to understand how other road users act, and, in turn, how we should drive in complex situations. The search acts as a safety net and ensures the robot navigates the environment safely, smoothly, and predictably."^[1]

The split is worth unpacking. The learned half is a prediction model: given camera and radar input, a neural network estimates what nearby pedestrians, cyclists and drivers are about to do and what a competent driver would do in response. Networks suit that job because road behavior is too varied to enumerate by hand, but they are statistical rather than guaranteed and can produce a confident output in a situation nothing in training resembles. The search half is classical motion planning: the planner enumerates and scores candidate trajectories against hard constraints such as collision-free geometry, drivable surface and speed limits, then picks one. It is slower and less flexible than an end-to-end learned policy, but its behavior can be bounded and reasoned about. Running search over the model's output keeps a network's occasional bad guess from becoming a bad maneuver, and it is the architectural family most deployed self-driving car stacks have converged on.

Training and validation

"We train with behavior cloning and reinforcement learning on large, diverse data so the system keeps improving with real-world experience."^[1] Behavioral cloning is a form of imitation learning in which the model is fit to recorded expert driving, learning to reproduce the action a human took in each observed situation. Reinforcement learning then optimizes against a reward signal, which lets the system improve on cases where simply copying a demonstration is not enough or where no demonstration exists.

Validation is described as testing "against a range of edge cases in diverse suburban environments, including construction zones with temporary signage, trucks blocking bike lanes, pedestrians in low light conditions, animals off leash in residential areas, broken traffic lights, school zones, and much more."^[1] Tang said Dot had been tested across millions of simulated and real-world miles; DoorDash's consumer page says "hundreds of thousands of miles on public roads."^[3][6] No disengagement, intervention or per-mile incident rate has been published.

In development

Two forward-looking items are the most technically interesting. The first is a foundation driving model: DoorDash says it is building a model that "handles the transition between road driving, sidewalk navigation, and precise driveway maneuvering, all within a single inference pass," built "to generalize, handling the infrastructure patterns, local driving behaviors, and environmental conditions that vary from location to location."^[1] Transitions between road, sidewalk and driveway contexts are normally handled by separate modes with handoff logic between them; collapsing them into one foundation model evaluated in a single inference pass removes those seams, and the emphasis on generalization across locations is a statement about how DoorDash intends to enter new cities without rebuilding the stack for each.

The second is reinforcement learning scaled to the delivery long tail: DoorDash says it is "scaling RL systems to handle the long tail of delivery-specific scenarios," giving as examples "navigating around a child's bike left in a driveway, or safely approaching a front door when a dog is wandering by," which it calls "scenarios that rarely appear in traditional AV datasets, but are daily realities for delivery."^[1] That is a real gap: driving datasets collected for robotaxis are dominated by road scenes, because robotaxis never leave the road.

The claimed data advantage

DoorDash's argument for why a delivery company can build competitive autonomy is data it says nobody else has: "we have billions of data points about effective merchant pickups and consumer drop-offs all in the last 10 feet, and billions of ground-truth commercial outcomes," meaning whether the delivery succeeded, whether the consumer was satisfied and whether the merchant sold more.^[1] Tang put it in business terms: "After more than 10 billion deliveries, we have data on what works, what breaks and what scales."^[6] The claim is untested. Handoff-outcome logs from human couriers are not driving data, and DoorDash has not explained how commercial-outcome labels enter the training of a driving policy.

How autonomous is Dot in practice?

This is the most consequential open question about Dot, and DoorDash's public statements do not fully agree.

At the September 2025 launch, Rege told TechCrunch that Dots cannot be remotely operated by humans, that DoorDash believes teleoperation is not the right approach for Dot, and that the robot is instead trained to wait and pull over to the side of the road, with a field operator dispatched to its location in serious circumstances.^[4] DoorDash's own consumer page, live in 2026, says something different: "If Dot encounters an issue, remote operators can step in to assist, and a local operations team is on hand to resolve anything that can't be managed remotely."^[3] Fremont's permit is more specific still, providing that in Phase 1B up to 30 robots may "operate autonomously within the Service Area and with the capability to be operated remotely as needed," after a Phase 1A period of roughly two months in which no more than three robots run and are "chaperoned by local operators."^[8][9] On the Q1 2026 earnings call, CEO Tony Xu listed remote operations as one of the pillars the company is hardening: "It is not just the autonomy. It is the autonomy, the hardware, the remote operations, all the work around regulatory with the different cities so that we can do this at scale."^[18] Bloomberg reported in September 2025 that remote support of the robots was already a staffed function at DoorDash Labs.^[5]

These statements do not fully reconcile. Dot drives with no human aboard and no chase vehicle, and Rege described a field-operator model rather than a teleoperation model, but Fremont's permit contemplates a remote-operation capability and DoorDash's own consumer page describes remote operators stepping in, which is hard to square with a flat statement that Dots cannot be remotely operated. DoorDash's own model statute for other states likewise defines a personal delivery device as one that "can operate with or without a person actively controlling it."^[16] What is clear either way is that Dot is not unsupervised: a remote operations function and a local field team both exist. DoorDash has not published an intervention rate, a supervision ratio, or a description of what remote operators may command. Arizona law makes some human oversight a condition of operating at all: the statutory definition of a personal delivery device requires "automated driving technology, including software and hardware, that enables the operation of the device with the remote support and supervision of a human," and a separate section requires a human agent "capable of monitoring or exercising physical control over the navigation and operation" of the device.^[14][15]

Deployment

Scroll sideways for more →

Dot is built in Fremont. DoorDash's own release says only "Manufactured in Fremont, Dot is an all-electric, compact robot"; the contract manufacturer, Sonic Manufacturing Technologies, is named by Chain Store Age and not by DoorDash.^[9][10] That answer arrived about five months after the unveiling, at which DoorDash declined to say where the robot was made and Rege told Bloomberg Television the company was still discussing manufacturing and component sourcing.^[5] DoorDash also declined to disclose fleet size at launch, has not disclosed it since, and declined to say how many Phoenix-area businesses were using Dot.^[5][11] No cumulative delivery count and no named participating merchant have been published; the only hard fleet figure anywhere is Fremont's permit ceiling of 30 robots.^[8]

Regulatory basis

In Arizona, Dot operates under the state's personal delivery device statute, Arizona Revised Statutes sections 28-1221 through 28-1228, enacted in 2020. A compliant device "is not a vehicle" for purposes of the traffic code; it may travel up to 12 mph in a pedestrian area (a local authority may lower that to no less than 7 mph) and up to 20 mph "on the side or shoulder of a highway in an area that is not a pedestrian area"; it must carry a marker with the owner's name, contact information and a unique identification number, a braking system and night lighting; and its operator must carry at least $100,000 of general liability insurance.^[13][14][15] The statute sets no weight limit at all, which is what leaves a 350 lb machine inside the definition; DoorDash's own model bill for other states proposes a 500 lb cap excluding cargo.^[16] That was not always so. Arizona's first personal delivery device law, enacted in 2018 as ARS 28-913, defined the devices as operating "primarily on sidewalks and within crosswalks," weighing "less than two hundred pounds, excluding cargo" and running at "a maximum speed of seven miles per hour," each subject to local authorization, and it repealed itself "from and after August 31, 2020."^[29] A 350 lb, 20 mph Dot would have fallen outside that definition; the 2020 replacement article dropped both limits.

California has no comparable statewide framework, and DoorDash does not appear on the California DMV's list of autonomous vehicle testing permit holders as of the DMV's 8 May 2026 update.^[23] Fremont instead authorized Dot through a municipal encroachment permit, which records what a city actually demands of a road-going delivery robot: phased robot counts, per-surface speed caps, a standardized Operating Plan and Incident Response Procedure, engagement with cycling and pedestrian groups and city police, fire and public works staff, a requirement to map the service area "to render a real-time 3D representation of the street environment," and revocation at any time for non-compliance, safety issues, "significant public concern," or direction by the City Manager or Council.^[8]

DoorDash is lobbying other states to adopt a model personal delivery device law. Its July 2026 deck for the New Mexico legislature proposes an identification number and company contact on each device, working brakes and lights, 10 mph near pedestrians and 20 mph elsewhere, a 500 lb weight cap excluding cargo, $100,000 of liability insurance, and a clause barring cities and towns from adding taxes, fees or rules except for public health and safety.^[16] State Senator Linda Trujillo called the proposed limit on local regulation a "huge red flag."^[17] DoorDash told the committee that state legislation is not technically required for it to deploy.^[17]

Safety record and incidents

Beyond the testing-mileage line above, DoorDash's only public safety statements are that Dot does not use facial recognition and that no driving data is connected to any customer or passerby.^[3]

One incident has been independently reported. On 15 June 2026, near Ray Road and Hamilton Street in Chandler, Arizona, a Dot making a delivery encountered a SWAT team responding to a weapons-related incident. Police ordered it to leave; it stayed in place, remaining on scene while officers detonated a flashbang through a window of the house. A DoorDash box truck arrived afterward and an employee manually piloted the robot away. A DoorDash spokesperson told The Independent that the robot "behaved as designed," "stopping and waiting safely while authorities managed the scene," and that the company was "grateful to the Chandler Police Department for their professionalism," adding, "We're continuing to review the incident and will share more as we learn it."^[20] The episode illustrates the failure mode Rege described in 2025: when a situation exceeds the policy, Dot stops rather than improvising, and a human comes to it.

Competitive position

Dot occupies a category with few members. Nuro's road-going delivery pods are the closest precedent: in February 2020 the National Highway Traffic Safety Administration granted Nuro a temporary exemption from certain low-speed vehicle standards for its occupantless R2, capped at 5,000 units over two years, a vehicle with no windshield or mirrors. Dot is regulated as a personal delivery device rather than as a motor vehicle, a materially lighter compliance path.^[21]

On the sidewalk side, Serve Robotics said in December 2025 that it had built more than 2,000 robots, which it called the largest sidewalk delivery fleet in the United States, operating at Level 4 across seven US markets with DoorDash and Uber Eats among its platform partners.^[22] Starship Technologies reports more than 10 million deliveries, 14 million miles driven and 300-plus locations, and describes its robots as "99% autonomous."^[24] Coco Robotics supplies sidewalk robots to DoorDash itself in Los Angeles, Chicago and Miami.^[10] Bloomberg noted that Dot's 20 mph exceeds the top speeds of both Coco's and Serve's machines.^[5]

DoorDash's framing is that these are complements rather than substitutes: the Autonomous Delivery Platform assigns each order to a Dasher, a Dot, a sidewalk robot or a drone in real time, and the engineering post positions the whole effort against the robotaxi industry: "While others aim to build end-to-end AI for driving, we're building end-to-end AI for commerce."^[1][2] Dot is the ground leg; drone delivery through Wing and Flytrex, and from July 2026 the in-house DoorDash Air program, is the air leg.

On the Q1 2026 earnings call, CFO Ravi Inukonda said only that DoorDash was "very happy with the progress" and had "launched it in a couple of markets." Xu added a caution that reads as a corrective to the launch coverage: "There is a pretty big difference between just shipping a vehicle or having a vehicle ready for a demonstration and a vehicle that can really operate at scale under any condition and is really battle-tested."^[18]

See also

• DoorDash
• DoorDash Labs
• Autonomous Delivery Platform
• DoorDash Air
• Delivery robot
• Autonomous driving
• Waymo
• Drone delivery

References

1. ^Stanley Tang and Ashu Rege, "Engineering Autonomy for Local Commerce: Building Dot and the Autonomous Delivery Platform," DoorDash Engineering Blog, 30 September 2025. careersatdoordash.com/...onomous-delivery-platform
2. ^"DoorDash Unveils Dot, the Delivery Robot Powered by its Autonomous Delivery Platform to Accelerate Local Commerce," DoorDash Newsroom, 30 September 2025. about.doordash.com/...doordash-unveils-dot
3. ^"Dot," DoorDash product page, accessed 31 July 2026. about.doordash.com/...dot
4. ^Maxwell Zeff, "DoorDash unveils Dot, its autonomous robot built to deliver your food," TechCrunch, 30 September 2025. techcrunch.com/...robot-built-to-deliver-your-food
5. ^"DoorDash Unveils Delivery Robot in Hardware Debut," Transport Topics (Bloomberg News), 30 September 2025. ttnews.com/...doordash-delivery-robot
6. ^Julie Littman, "Why DoorDash built its own delivery robot," Restaurant Dive, 1 October 2025. restaurantdive.com/...761494
7. ^"DoorDash Launches The Cutest 20mph Delivery Robot," Ride AI newsletter, 30 September 2025. rideai.org/...ches-the-cutest-20mph-delivery-robot
8. ^"DoorDash Delivery Robots," City of Fremont, California, accessed 31 July 2026. fremont.gov/...doordash-delivery-robots
9. ^"DoorDash and City of Fremont Autonomous Delivery Collaboration," DoorDash Newsroom, 2 March 2026. about.doordash.com/...doordash-fremont-collaboration
10. ^"DoorDash brings proprietary delivery robot to new market," Chain Store Age, 16 March 2026. chainstoreage.com/...ary-delivery-robot-new-market
11. ^"DoorDash rolls out autonomous delivery robots in East Valley, plans wide expansion," Arizona Technology Council, reprinting Phoenix Business Journal, October 2025. aztechcouncil.org/...-out-robots-in-tempe-and-mesa
12. ^"Autonomous Robot To Begin Making Deliveries In Riverside," Patch, 27 October 2025. patch.com/...bot-begin-making-deliveries-riverside
13. ^Arizona Revised Statutes 28-1225, "Areas of operation; speed limits," Arizona State Legislature, accessed 31 July 2026. azleg.gov/...01225
14. ^Arizona Revised Statutes 28-101, "Definitions" (definition of personal delivery device), Arizona State Legislature, accessed 31 July 2026. azleg.gov/...00101
15. ^Arizona Revised Statutes 28-1223, "Operator," Arizona State Legislature, accessed 31 July 2026. azleg.gov/...01223
16. ^"Building the Future of Autonomous Delivery," DoorDash presentation to the New Mexico Legislature Transportation and Infrastructure Revenue Subcommittee, 14 July 2026. nmlegis.gov/...%20071426%20Item%205%20DoorDash.pdf
17. ^"Delivery company DoorDash wants to roll out robots on New Mexico roads," Source New Mexico, 15 July 2026. sourcenm.com/...oll-out-robots-on-new-mexico-roads
18. ^"DoorDash (DASH) Q1 2026 Earnings Call Transcript," The Motley Fool, 6 May 2026. fool.com/...rdash-dash-q1-2026-earnings-transcript
19. ^"DoorDash is building its own drone delivery business," TechCrunch, 29 July 2026. techcrunch.com/...-its-own-drone-delivery-business
20. ^Frank Landymore, "DoorDash Delivery Robot Invades SWAT Scene, Won't Leave as Cops Flashbang Resident," Futurism, 24 June 2026. futurism.com/...delivery-robot-swat-scene
21. ^"NHTSA Grants Nuro Exemption Petition for Low-Speed Driverless Vehicle," National Highway Traffic Safety Administration, 6 February 2020. nhtsa.gov/...petition-low-speed-driverless-vehicle
22. ^"Serve Robotics Builds 2,000 Autonomous Delivery Robots, Creating Largest Sidewalk Delivery Fleet in the U.S.," Serve Robotics via GlobeNewswire, 12 December 2025. globenewswire.com/...alk-Delivery-Fleet-in-the-U-S
23. ^"Autonomous Vehicle Permit Holders," California Department of Motor Vehicles, testing permit list as of 8 May 2026, accessed 31 July 2026. dmv.ca.gov/...omous-vehicle-testing-permit-holders
24. ^Starship Technologies company page, accessed 31 July 2026. starship.xyz/company
25. ^Harrison Shih et al., "Automated vehicle for autonomous last-mile deliveries," US Patent 11,372,426 B2, assigned to DoorDash Inc., filed 5 December 2019, granted 28 June 2022. patents.google.com/...en
26. ^Harrison Shih, Ilya Polyakov and Bruce Vest, "Automated delivery vehicle," US Design Patent D931,756 S, assigned to DoorDash Inc., filed 17 May 2019, granted 28 September 2021. patents.google.com/...en
27. ^Harrison Shih, Ilya Polyakov and Bruce Vest, "Automated delivery vehicle," US Design Patent D932,368 S, assigned to DoorDash Inc., filed 17 May 2019, granted 5 October 2021. patents.google.com/...en
28. ^"Introducing DoorDash Labs: DoorDash's Robotics and Automation Arm," DoorDash Newsroom, 18 November 2021. about.doordash.com/...-robotics-and-automation-arm
29. ^House Bill 2422, Arizona 53rd Legislature, 2nd Regular Session (2018), adding and delay-repealing ARS 28-913. azleg.gov/...HB2422C.pdf