[Future of Delivery] Ordering from a Chicken Restaurant 800m Away... Delivery Robot Rings 'Ddingdong' in 15 Minutes

Average speed 5-6 km/h... Not much different from typical adult walking speed
Passersby appear, route changes naturally... Recognizes pets too
High-cost LiDAR sensor replaced with camera... "Development cost about one-tenth"
Traffic light distance requires control center intervention... Delivery to door likely to take time

Delivery robot 'NewBee' by Newbility, currently being demonstrated in Songdo, Incheon.
Photo by Joonhyung Lee

[Asia Economy Reporter Junhyung Lee] On the 27th, at Yonsei University International Campus Jinri Building D in Songdo, Incheon. When an order was placed at a chicken restaurant about 800 meters away, the cooked chicken was loaded onto a delivery robot via a conveyor belt. The delivery robot carrying the chicken headed straight to its destination. The average speed was 5 to 6 km/h, not much different from the average walking speed of an adult (4 km/h). When the delivery robot encountered a crosswalk on the way, it stopped and waited for the green signal. After about 15 minutes, the delivery robot arrived at the destination and sent a notification to the customer application (app). After completing the delivery, the robot returned to the chicken restaurant along the same route.

Delivery robot startup Newbility conducted a demonstration on this day ahead of its pilot service. Newbility’s AI delivery robot ‘Newbee’ operated stably without slipping even on rainy roads. Puddles formed in various places on the street were not a problem. When a pedestrian appeared during operation, it naturally changed its route to go around. Lee Sangmin, CEO of Newbility, said, “We have conducted tests in much harsher environments than other companies,” adding, “It can drive without any issues on roads commonly seen in cities.”

'Newbee' with One-Tenth the Price... Replacing LiDAR with Cameras

The approximately ten cameras installed on Newbee are one of Newbility’s core technologies. Newbee collects surrounding information through high-performance cameras such as omnidirectional cameras capable of 360-degree shooting and stereo cameras that capture two images simultaneously. Newbility replaced the expensive LiDAR sensors used in existing delivery robots with cameras, significantly lowering the robot’s unit cost. LiDAR sensors cost several million won each, making the typical price of delivery robots at least tens of millions of won. The robot platform applied to Baedal Minjok’s delivery robot ‘Dilly’ is known to cost about 48 million won per unit. In contrast, Newbee’s development cost is only about 5 million won.

The company focused on developing Visual SLAM to avoid using LiDAR sensors. Visual SLAM is a technology that uses cameras to determine location while simultaneously creating a virtual map reflecting the surrounding environment. CEO Lee said, “The technology to properly position cameras and apply them to AI is considered unique in the industry.”

The delivery robot 'Newbee' operating at Yonsei University International Campus located in Songdo, Incheon. When a pedestrian appeared, it naturally changed its route and turned away.
[Photo by Joonhyung Lee]

In practice, Newbee immediately stopped when a vehicle appeared while crossing a crosswalk without traffic lights. It then waited for the vehicle to pass and resumed operation after confirming that obstacles had cleared. When there were no pedestrians on the sidewalk, it positioned itself in the center of the path. This was designed so that the AI balances the width of the road on both sides to avoid the risk of falling. Newbee also recognized low-height pets through cameras installed on its underside.

Technical Limitations Remain... Cannot Make Independent Decisions at Traffic Lights

The company is also building big data for stable AI driving. Since early this year, Newbility has tested dozens of delivery robots in Songdo and other areas, collecting 8,000 to 9,000 images of driving environments daily. Employees, including CEO Lee, have devoted time to annotation (data processing) work whenever possible. CEO Lee explained, “Since sidewalk environments differ by region, datasets for each area become key to autonomous driving quality,” adding, “The winner in delivery robots will be determined by how much data is secured across various environments.”

The delivery robot 'Newbee,' currently being tested in Songdo, Incheon, is crossing the pedestrian crosswalk.
Photo by Joonhyung Lee

However, there were technical limitations. Newbee could not make independent decisions when encountering traffic lights during operation. For safety reasons, the operation authority at locations with traffic lights is currently transferred to the control center. The control center assesses the surrounding environment through Newbee’s cameras and, if it determines that AI can drive alone, returns the operation authority to the AI. After regaining operation authority, Newbee stopped at a crosswalk that takes about 10 seconds to walk across when the green light had only 6 seconds left and waited for the next signal. This is because it is set to cross the crosswalk only when the traffic light changes from red to green. CEO Lee said, “The control center’s intervention time at traffic light locations is about 5 seconds,” adding, “As driving data accumulates, the need for the control center will decrease.”

Delivery to the doorstep is also expected to take time. Newbee can currently only deliver up to the building entrance. Due to technical limitations, indoor operation such as opening shared entrance doors and using elevators is restricted. A Newbility official said, “Delivery to individual units is a task currently underway,” adding, “Considering this, we plan to provide incentives such as points when using the delivery robot.”