RIVR: Bringing Physical AI to Last-Mile and Doorstep Delivery

RIVR: Bringing Physical AI to Last-Mile and Doorstep Delivery

Cities are hostile terrain for automation. Sidewalks are uneven, environments change constantly, and human behavior is unpredictable. RIVR approaches this challenge with a focus on Physical AI—robotic systems designed to learn, adapt, and operate in the physical world. Founded as a spin-out from ETH Zurich’s Robotic Systems Lab, RIVR focuses on automating the “last mile” and “last 100 yards” of deliveries—tasks that involve navigating stairs, curbs, and cluttered city environments.

Technology and Product

RIVR’s platform combines AI software with a hybrid wheeled-legged robot design, sensor systems, and fleet management infrastructure. According to the company, its Physical AI architecture relies on reinforcement learning and supervised learning from simulated and real-world data to enable robots to perceive their surroundings, make decisions, and act autonomously in dynamic environments. The robots use wheels for efficient travel and articulated legs to handle uneven terrain, stairs, and obstacles such as curbs—features RIVR positions as critical for urban delivery. Cameras and LiDAR support navigation and obstacle avoidance, while cloud-based monitoring enables remote supervision and intervention when required. Product configurations include both fully autonomous operation and van-assisted delivery, where robots work alongside human drivers to increase delivery throughput.

RIVR aims to revolutionize last-100-yard parcel delivery for e-commerce, logistics and retail. (© RIVR)

Industrial Fit and Applications

The startup targets logistics providers, retailers, e-commerce platforms, and food delivery services looking to address rising delivery volumes, labour constraints, and cost pressure in cities. The robots are designed to operate in mixed urban environments, including sidewalks, crossings, and densely populated areas where earlier delivery robots faced limitations. Reported pilot deployments include urban trials in cities such as Zurich, where robots have been tested for food and parcel delivery in collaboration with logistics and delivery partners. RIVR frames these deployments as opportunities to gather operational data, which it argues can be used to continuously improve robot performance and robustness over time

The trial-and-error learning approach uses simulated data and GPU simulations to train robots efficiently. (© RIVR)

Founding Team

RIVR was founded by Marko Bjelonic (CEO), Lorenz Wellhausen (CTO Software), Giorgio Valsecchi (CTO Hardware), and Alexander Reske (COO). The founding team combines backgrounds in robotics research, AI development, and systems engineering, drawing on academic experience in autonomous systems to build commercially deployable delivery robots.