How Serve Robotics Achieved 99.8% Success for Last-Mile Autonomous Delivery
Serve Robotics · United States
Serve Robotics, which spun off from Uber in 2021, operates over 1,000 physical AI-powered sidewalk delivery robots serving over 2,500 restaurants across five cities. Its third-generation robots are simulated in NVIDIA Isaac Sim and powered by NVIDIA Jetson Orin edge AI platforms, achieving a 5x compute improvement over the prior Xavier generation and 12+ hours of battery life. The fleet has completed over 100,000 autonomous deliveries with a 99.8% completion rate, logging about 1 million miles of data monthly (nearly 170 billion image-LiDAR samples) to improve navigation and HD maps. The company plans to deploy 2,000 robots by the end of 2025.
Overview
Serve Robotics, which spun off from Uber in 2021, operates over 1,000 physical AI-powered sidewalk delivery robots serving over 2,500 restaurants across five cities. Its third-generation robots are simulated in NVIDIA Isaac Sim and powered by NVIDIA Jetson Orin edge AI platforms, achieving a 5x compute improvement over the prior Xavier generation and 12+ hours of battery life. The fleet has completed over 100,000 autonomous deliveries with a 99.8% completion rate, logging about 1 million miles of data monthly (nearly 170 billion image-LiDAR samples) to improve navigation and HD maps. The company plans to deploy 2,000 robots by the end of 2025.
The challenge
The last-mile delivery industry has long struggled with inefficiency and rising costs, with traditional delivery methods relying on two-ton cars to transport two-pound meals, creating unnecessary traffic congestion and emissions. Serve Robotics also faced CPU compute limitations that prevented real-time autonomy, and power consumption constraints that impacted battery life and made all-day field operations difficult to achieve.
The solution
Serve Robotics partnered with NVIDIA to implement a comprehensive robotics development platform. It used NVIDIA Isaac Sim, built on NVIDIA Omniverse, to validate autonomous performance at scale through thousands of simulated scenarios and synthetic data generation, testing edge cases before real-world deployment. For hardware, Serve implemented NVIDIA Jetson platforms, progressing from Xavier to the more powerful Orin AGX as its robots evolved, enhanced with Jetpack development tools and TensorRT inference optimization.
Reported business value
Serve's third-generation robots gained a 5x compute boost moving from Xavier to Orin, with 12+ hours of battery life per charge. The fleet has completed over 100,000 autonomous deliveries with a 99.8% completion rate, averaging over 40% quarter-over-quarter growth since Q1 2022, and now serves 2,500+ restaurants across five cities, logging about 1 million miles of data monthly, nearly 170 billion image-LiDAR samples, to continuously improve robot learning, HD maps and navigation.
Sources
Open any source and check the claim yourself — that is the point of the register.
Other logistics entries in the register.
ID Logistics' SmartVision delivers 80% error reduction in warehouse order picking
ID Logistics launched SmartVision, an AI solution using stereovision cameras to interpret real-time data and monitor parcel preparation by warehouse operators. The system reduces picking error rates by 80% by continuously recording actions on the trolley and validating the number of parcels deposited in real time; recorded images also serve as verification for complaints. Team leaders get real-time visibility into quality rate, remaining volume, and palletizing conformity. The solution was first deployed at a food distribution site north of Paris, with plans to extend to retail sites in France and internationally.
PostNL embeds AI-driven route planning and sales intelligence across its logistics platform
PostNL's 2025 transformation embedded artificial intelligence and advanced analytics throughout operations: predictive algorithms support real-time route planning, parcel flow optimisation and capacity forecasting, and a new AI-powered sales intelligence tool converts ideal customer profiles into prioritised prospect lists with detailed shipment insights for account managers. In Belgium, PostNL piloted a fully digital onboarding journey for small business customers, reducing registration time from 15-20 days to a maximum of one day.
Delhivery achieves 160 ms latency for high-precision geocoding using Amazon EKS
Delhivery, a logistics provider in India, implemented a fine-tuned open-source Llama 3.2 1B large language model on Amazon EKS to support high-volume geocoding of pickup and drop-off addresses. The system processes up to 8,000 requests per minute at 160 milliseconds latency using NVIDIA A10G GPU-backed G5 Xlarge instances and the vLLM framework. Delhivery cut model-serving costs by approximately 80 percent and accelerated prototyping cycles from two days to under six hours, working with the AWS Prototyping and Cloud Engineering (PACE) team.
C.H. Robinson Scales Fleet of AI Agents Past 30
C.H. Robinson has deployed more than 30 proprietary AI agents at scale across its global logistics network, spanning the shipment lifecycle from quoting to tracking. The agents have automated over 3 million shipping tasks and contributed to a 30% increase in productivity since 2023. Named agents include a Quoting Agent (32-second price quotes, 1M+ quotes), an Orders Agent (processes 5,500 truckload orders/day, saving 600 labor hours daily), an LTL Classifier Agent (75%+ of LTL orders automated, up from 50%), an Appointments Agent (3,000+ appointments/day across 43,000 locations), a Truck Posting Agent, and a Shipment Tracking Agent.
Was this helpful?
Your feedback helps us improve our use case database