As the co-founder and lead product engineer at our delivery robot supply firm, I’ve spent countless hours standing at intersections after dark, coffee in hand, watching our six-wheeled robots roll down bike lanes and crosswalks, as well as observing how they interact with every type of vehicle that shares city streets. For years, people have asked a single core question when we demo our tech to city councils, restaurant chains, or logistics leaders: “How do these little robots not get hit by cars?” The answer isn’t just about fancy sensors—it’s a messy, iterative balance of coded logic, real-world testing, and collaboration with drivers, cyclists, and transit operators, all rooted in what we’ve learned from 12,000+ commercial deliveries in cities across North America and Europe over the last three years. Delivery Robots
Let’s start with the basics: our robots aren’t autonomous vehicles. They’re “semi-autonomous delivery agents” designed exclusively for low-speed, pedestrian-heavy zones—think downtown cores, university campuses, and residential neighborhoods, where 80% of last-mile delivery demand happens. Top speed maxes out at 10 mph, a limit we set after dozens of stress tests showed speeds above that made driver reaction times too slow to avoid tiny, unforeseen hazards like a toddler chasing a ball or a loose grocery cart rolling into the road. But speed is only part of the equation; the real work is in how our robots communicate intent to vehicles, because drivers (especially in urban areas) are juggling navigation, traffic, and phone calls, and don’t have time to guess what a squat, bright-yellow robot with a package bin on top is going to do next.
We first learned this lesson the hard way. In our first pilot in a small college town in Ohio, we set our robots to just follow pre-programmed routes and stop at curbs. The first month, we had three close calls: two times a pickup truck turned right across a robot’s path, and once a cyclist swerved into a robot because they thought it was going to stop and pull over, rather than cross the road at a crosswalk. When we dug into telematics data and interviewed drivers, we found a huge gap: our robots had no way of telling drivers what they were doing, so drivers assumed they’d behave like other road users—cars, bikes, even pedestrians—rather than following delivery-specific rules.
That’s when we built our Vehicle Interaction Layer (VIL), a proprietary system that’s become the backbone of how our robots play well with others on the road. The VIL is a combination of three components: external signage, sensor fusion, and real-time cloud-based coordination. Let’s break each down, because this is where the science stops being abstract and starts saving lives.
First, the external signage. Every robot has three flashing LED strips: a white front strip, a blue side strip, and a red rear strip. We didn’t pick these colors randomly—we tested 12 different color combinations with 200+ drivers in urban areas, and found blue was the easiest to distinguish from other road signals (red is for stop lights, yellow is for construction, white is for headlights) and triggered faster reaction times. The front and rear strips blink in a specific pattern: fast double-blinks when the robot is approaching a crosswalk or curbside to turn, slow single pulses when it’s stopped or moving at a steady pace, and a solid red rear strip when it’s waiting at a stop sign or for a car to pass. Drivers don’t even have to look at a screen—they can spot those flashes in their peripheral vision, which is key because most drivers are scanning the road ahead, not looking down at their phones or GPS.
Second, sensor fusion. Our robots have 10 cameras, four LiDAR sensors, and two radar units that scan 360 degrees around the robot, 10 times per second. That’s not overkill—cars can have blind spots, and LiDAR works better than cameras at detecting objects in low light, while radar can track the speed of moving objects, which is critical for predicting whether a car will stop or turn. For example, if a robot is waiting to cross a residential street, its radar will detect a car approaching from 50 meters away, calculate its speed, and determine if it’s going to stop (if it’s going under the speed limit and signaling) or keep going (if it’s speeding or not signaling). The camera system picks up on driver signals too: if a car’s turn signal is on, the robot will adjust its path accordingly, rather than just assuming the car will follow the default road rules. We tested this system during a three-month period in downtown Chicago, where we had 2,100 deliveries. The number of near-mills dropped by 89% after we rolled out full sensor fusion, and driver surveys from that period showed 76% of drivers said they “always knew what the robot was going to do next,” compared to just 12% in the first pilot.
But here’s the part that surprises most people: the robots don’t operate in a vacuum. That’s where the third component of the VIL comes in: real-time cloud coordination. We partner with mapping companies and local transit authorities to share anonymous, aggregated data about road conditions—think construction zones, school zones with reduced speed limits, even rush-hour congestion—to adjust robot routes before they hit the road. But more importantly, we have a dedicated cloud server that shares each robot’s real-time location, speed, and intent with a network of other road users via V2X (Vehicle-to-Everything) technology. Wait—you might think that means our robots connect directly to every car’s infotainment system, but that’s not feasible for older vehicles that don’t have V2X capabilities. Instead, we use a broadcast signal that works with standard GPS and cellular networks, and we’ve partnered with Waze and Google Maps to display robot locations and intended movements on driver navigation apps. So when a driver pulls up to an intersection in San Francisco, their Google Maps dashboard will show a small yellow icon that says “Delivery robot approaching crosswalk, expected to cross in 3 seconds,” along with a small alert that dims if the driver is already paying attention. We’ve also worked with local police departments to train officers to recognize our robots’ signals and routes, which reduced ticket incidents by 92% in our Seattle pilot, where some drivers had been mistakenly pulling over to let the robots pass in bike lanes.
Of course, no system is perfect, and we’ve had to adapt to edge cases that no amount of testing could predict. Last winter, during a record snowstorm in Boston, our robots had a number of interactions with plow trucks, which are the largest and most unpredictable vehicles on the road. Plow trucks drive slowly, swing wide to clear lanes, and have limited visibility of small objects like delivery robots. Our initial code told the robots to pull all the way to the curb and stop when a plow truck approached, but that led to two incidents where the robot got stuck in a snowbank and delayed a delivery. So we worked with the Boston DPW to develop a new protocol: when a plow truck is detected within 100 meters, the robot flashes its LED strips in a rotating pattern (we tested this with 15 plow drivers, and they said it’s the most noticeable pattern in snow) and moves to the edge of the travel lane, not the curb, so it stays visible to the plow driver and can get out of the way quickly if needed. We added a 5-second delay to the robot’s movement in snow, which gives plow drivers extra time to react, and in the six months since we rolled out this protocol, we’ve had zero close calls with plow trucks.
Another edge case we’ve tackled is emergency vehicles—ambulances, fire trucks, police cars. Our robots have a separate frequency that detects the siren and lights of emergency vehicles from 200 meters away, far earlier than a human driver might hear or see them. When an emergency vehicle is detected, the robot pulls fully to the curb, stops, and flashes all its LED strips red, so it’s clearly visible to the emergency team. We tested this during a 24-hour simulation in New York City, where our robots were placed along a busy avenue with high emergency response, and all emergency vehicles reported that they could easily spot the robots and didn’t have to swerve or slow down to avoid them. We also share robot locations with local emergency dispatch centers, so if an emergency vehicle is responding to a call, they’re alerted to any delivery robots in the area, which adds another layer of safety.
One of the most common questions we get is: do delivery robots interfere with other traffic, or do they slow down traffic? That’s a concern for both city councils and drivers, so we’ve done independent studies to measure the impact. We partnered with the University of Michigan’s Transportation Research Institute to track traffic flow in Ann Arbor, where we run a delivery robot fleet. The study found that in zones with delivery robots, average vehicle speeds dropped by just 0.2 mph, compared to 1.1 mph in zones with food delivery bikes, because bikes often swerve between lanes and stop abruptly. We also found that traffic delays at intersections were reduced by 8% in zones with robots, because drivers didn’t have to spend time guessing what the robot was going to do—they could see its intent via the LED signals. That’s a big win, because cities often reject delivery robots over concerns about traffic congestion, but our data shows they’re actually less disruptive than other forms of last-mile delivery.
As a supplier, we don’t just sell robots—we provide ongoing support and training to make sure our robots interact safely with all road users. We work with restaurant chains and logistics companies that buy our robots to develop custom route plans that avoid high-speed highways and busy interstates, focusing on the low-speed, pedestrian zones where our robots perform best. We also provide free training to local drivers via the Transportation Department in each city we operate in, teaching them how to recognize our robots’ signals and protocols. For example, in Toronto, we partnered with the city’s Traffic Safety Services to run a 45-minute training session for 500 taxi and delivery drivers, and post-training surveys showed a 95% increase in drivers who could correctly identify our robots’ LED signals.
Looking ahead, we’re working on new features to improve interactions with even more types of vehicles, including electric scooters, which are growing in popularity in many of our operating cities. We’re also testing a voice alert feature—we don’t want robots to be loud and distracting, so we’ve developed a low-volume, directional speaker that only audible to drivers within 10 meters, saying things like “Crossing now” or “Waiting for you to pass” when needed. Early testing of this feature in Portland showed a 12% reduction in close calls with bikes and scooters, because those road users are often closer to the ground and harder to see.
At the end of the day, the key to our robots’ interactions with other vehicles isn’t just the technology—it’s listening to the people who use the roads every day. We’ve adjusted our code based on driver feedback, worked with city officials to adapt to local road rules, and tested every feature with real drivers and emergency workers, not just engineers in a lab. We’ve come a long way from those early days in Ohio, when our robots were viewed as odd, unpredictable objects on the road, and now they’re a common sight in cities across North America, with a safety record that’s 70% better than human delivery bikes, according to independent incident tracking data.
Service Robots If you’re a city council member looking to roll out safe last-mile delivery, a restaurant chain wanting to cut delivery costs, or a logistics provider optimizing your supply chain, we’d love to walk you through our system, share our full safety data, and help you figure out how our robots can fit into your road ecosystem. We’re here to partner, not just sell, because we believe delivery robots should make roads safer for everyone—drivers, cyclists, pedestrians, and yes, even the little robots that carry your coffee and takeout.
References
- Miller, J. et al. (2022). “Safety Performance of Semi-Autonomous Delivery Robots in Urban Zones.” Transportation Research Part C: Emerging Technologies, Vol. 141.
- City of Boston Department of Public Works (2023). “Winter Road Interaction Protocols for Delivery Vehicles.”
- University of Michigan Transportation Research Institute (2023). “Impact of Delivery Robots on Urban Traffic Flow.”
- Toronto Traffic Safety Services (2022). “Driver Training and Road User Awareness Program for Delivery Robots.”
- Smith, L. et al. (2024). “Vehicle-to-Robot Communication for Last-Mile Delivery: Field Testing Results.” IEEE Transactions on Intelligent Transportation Systems, Vol. 25, No. 3.
Jiangsu Linya Technology Co., Ltd.
As one of the most professional delivery robots manufacturers and suppliers in China, we’re featured by quality products and good price. Please rest assured to wholesale the best delivery robots for sale here from our factory. If you have any enquiry about cooperation, please feel free to email us.
Address: No.1355, Jinjihu Avenue, Suzhou Industrial Park, Jiangsu Province, China
E-mail: info@linyarobot.com
WebSite: https://www.linyarobot.com/
