Summary
Cornell Tech is located in the southern part of Roosevelt Island, bordered by residential neighborhoods to the north and Southpoint Park to the south, with the East River on both sides. While the island has a few restaurants and a grocery store along Main Street, the limited variety and pricing often require students to travel off-island by tram or subway, especially after campus dining closes in the evening. At the same time, Main Street vendors discard unsold food daily, and this case explores how a Level 4 autonomous robot delivery system could bridge this gap by redistributing surplus food to students.
AV Use Case
What AVs are involved?
This solution involves the use of six Level 4 autonomous electric sidewalk delivery robots, which have the following characteristics: they are small and four-wheeled; their height is about 50 cm, they have an ability to carry cargo weight up to 20 kg; they can handle food cargo and have insulated as well as potentially refrigerated cargo compartments. These robots have LiDAR, 360° cameras, ultrasonic obstacle detection; they travel at a speed of 6 km/h on sidewalks, sharing pedestrian infrastructure. In addition to that, one Level 4 autonomous electric cargo van will work in concert with these robots on Main Street, picking up larger quantities of food from vendors with substantial leftovers at a speed of 35 km/h.
What are they doing?
The service runs every day from 3 pm to 10 pm, which is the time frame when there is excess food generated by the cafes, lunch counters, and local grocery stores just prior to their closing hours. The partner vendors put up surplus inventory information through a platform app that will feature their rescue pack, where the rescue pack will be sold 30 to 50 percent lower than the actual retail price. Then, the student can order a rescue pack through an app two hours before closing hours. The robot picks up the rescue pack of the vendor along Main Street following a predetermined route and delivers the goods to the doorsteps of the student after 25 minutes. One robot can service between two to three partner vendors and complete its delivery task within 25 minutes. The service launches as a twelve-month pilot, operating within existing sidewalk infrastructure and RIOC permit frameworks, with a go/no-go funding review at the six-month mark.
Why here?
The university whose curriculum teaches its engineering graduates how to design the next-generation cities stands on an island with limited options for shopping. In this respect, the challenges associated with obtaining food can be felt by students, many of whom are international, many of whom get funded and many of whom stay awake late into the night. The route between the street-vendor area on Main Street and the university campus is not long enough to render the idea of robotic delivery impractical but sufficiently remote. It is within this route where there is a concentration of all necessary elements - vendor oversupply and consumer demand, within the confines of one road network.
Stakeholders
Who participates?
The app and the partnering vendor arrangements, along with the marketplace for demand, are provided by the operator of the surplus food redistribution platform, which could be a start-up or an existing organization within the food rescue business sector. Cornell Tech will run the campus center and will manage student communication efforts regarding the program. Roosevelt Island Operating Corporation (RIOC) will provide sidewalk permits for the robots to operate on Main Street sidewalks and allow curb access at the vendor locations for pick-ups. Four to six Main Street vendors, comprising restaurants and other food providers including a café, lunch counter, and convenience store, will volunteer to take part, setting their own prices for the surplus on the platform and making rescue packs for the robots to collect. The financing of the first batch of six robots and the hub on campus is designed as a joint cost-sharing experiment in deploying new technology: Cornell Tech provides seed money via its budget for urban technology research, and the platform provider offers the software backbone in return for exclusivity over data collection on Roosevelt Island.
Who is impacted?
Late-working graduate students, the primary users, get access to an affordable meal without disrupting a work session to go elsewhere. International students, who may encounter even greater challenges with maneuvering their way after dark, receive an even greater benefit. Vendors can recover some amount of their income from food that would otherwise have been wasted, and any savings makes a difference when trying to run a small-scale food enterprise on an island that lacks enough customers for lunch hours. The elderly residents of Rivercross and Westview, which are residential towers immediately north of campus, face the same need for evening access to food as students. It does not appear that there will be any significant displacement effect on gig workers, given that the robots fulfill the surplus food redistribution mission and are not within the scope of current delivery services.
How does the solution use their capabilities?
The Cornell Tech contribution is its research facilities by integrating sensor-based data gathering within the robot fleet, which supports urban logistics research through a live platform. The operator of the platform provides vendor onboarding services, pricing algorithm, and order management services, which it has built up within pedestrian pick-up systems and simply needs to be adapted to autonomous last-mile delivery. The Main Street vendors’ contribution is their experience regarding surplus generation during the day, which helps the platform estimate window periods correctly and minimize idle periods for the robots. Finally, RIOC can streamline permitting at the island level, reducing the need for multiple approvals from New York City Department of Transportation that would typically be required elsewhere in the city.
How does it address their concerns?
The problem of food safety is resolved through the use of insulated and refrigerated holding areas in the robots that keep food at safe temperatures, with a strict maximum of 25 minutes for transit between vendor and student enforced by the routing algorithm. The food allergy problem is resolved by delivering digital ingredient labels for each order through the use of the app, which are obtained directly from the vendor’s inputs when preparing the rescue pack. The issue for vendors concerned with associating their brands with the discounted products is resolved by positioning the rescue pack as a sustainability initiative rather than a discount. Pedestrian safety is ensured by mandating a maximum speed limit for the robots on Main Street (6 km/h), as well as yielding behavior at all intersections. This, coupled with the small width of the robots, allows them to safely co-exist with two pedestrians at once on one standard pedestrian pathway. In addition, the routing algorithm is programmed to include one delivery zone north of Cornell Tech, specifically the Westview and Rivercross lobby areas, at no additional service cost. Pricing for elderly residents on fixed incomes is subsidized at cost by the platform operator as a condition of RIOC’s sidewalk permit, ensuring that the service does not function as an amenity exclusive to graduate students with institutional stipends.
Relevant Blueprints for Autonomous Urbanism
The following urban design strategies are drawn from the NACTO Blueprint for Autonomous Urbanism, 2nd Edition.
Human-Scaled Freight
The Blueprint suggests the consolidation and reduction of freight in size as a key direction for autonomous urbanism: deliveries must be consolidated in order to boost efficiencies, while vehicles must be reduced in size to accommodate their operational scales. The food rescue circuit works on this idea. Six electric sidewalk robots transport food from vendors to campus buildings over the final mile, traveling slowly like pedestrians and taking no more space on the sidewalks than one person would, carrying a bag. One cargo van serves the function of bulk transport whenever the volume of surplus goods in one place exceeds the capacity of a robot, supplementing rather than defining the system. This circuit is scaled to the corridor: narrow enough to walk alongside, slow enough to defer, and silent enough to function next to homes even at night.
The Challenge of Micro-Freight Devices
This problem of uncontrolled growth in the number of delivery robots on the sidewalks is identified in the Blueprint as the urban failure scenario, where the robots occupy space instead of being in harmony with pedestrians and cyclists. The deployment model considers this problem as the primary design criterion rather than an additional one. Every robot operates on the outer edges of the sidewalk, thus allowing pedestrians to walk freely through its middle. It also yields to any pedestrians at intersections. The dimensions of the robot allow two pedestrians to bypass it while remaining on the sidewalk itself. At the locations where deliveries are picked up from the vendors, the robot occupies a defined loading area for 90 seconds only and then vacates the space before the arrival of the next pedestrian group. After work hours, all robots retreat to their campus base, leaving the entire sidewalk space exclusively for pedestrians.
Methods
Step 1
- Tool: Google Maps
- Transformation: Utilized satellite maps and street-level images to map out the path connecting the Main Street vendors’ location and the Cornell Tech Campus buildings, determine potential pick-up locations, measure sidewalk width and crossing opportunities along the robot’s path, and analyze the circuit efficiency. Switched back and forth between the map and street views to ensure that there are no breaks in the sidewalk surface along the route and that each vendor entry point has sufficient curb-side parking space for the delivery robot.
- Result: After ensuring that the distance from Main Street to campus is sufficiently short for the sidewalk robot to make a round trip during one operational period, the four to six vendor locations for pickups were determined, along with the point at which the robot enters the campus and where deliveries take place inside the buildings.
Step 2
- Tool: ChatGPT (DALL-E)
- Transformation: Concept imagery created using prompts such as “A small autonomous robot with four wheels, picking up a paper bag of food from a cafe attendant on a narrow sidewalk within an urban setting, late afternoon warmth, modern glass university campus building in the background, photorealistic style, widescreen landscape view” and “A graduate student accessing the cargo bay of a small autonomous robot delivering a package near a modern glass university building entryway, evening lighting, contemporary university campus architecture, photorealistic style.” Repeated each prompt to better establish the robot’s scale and more clearly depict the human-robot interaction on both ends of the transaction.
- Result: Produced concept images depicting the vendor handoff on Main Street and the student collection at the campus building entrance.
Step 3
- Tool: Sora (OpenAI)
- Transformation: Used the generated images as references for Sora and asked it to create a small clip showcasing the food delivery process: an autonomous robot picking up food from a street vendor in the evening, followed by moving through an empty urban road and heading toward the glass building of a university.
- Result: Produced a video showing the robot’s journey from Main Street vendor pickup to campus delivery, demonstrating the end-to-end service experience.