Chinese autonomous delivery startup BemiGo introduced its full-stack technology platform at the Magic X Global Embodied Intelligence Innovation Conference on April 28 in Silicon Valley, marking the company’s first major showcase outside China. CEO Luke Chen used a keynote address to unveil three core products: a modular delivery platform, the SmartVLA end-to-end large model, and a cloud-based autonomy system called Brain 2.0. The company also detailed specifications for its T6 delivery robot, its flagship commercial vehicle.
Highlights
- T6 delivery robot offers 6-cubic-meter cargo volume, 1,000 kg payload, 200 km range, and 60 km/h top speed
- SmartVLA model deployed on NVIDIA Jetson Thor, marking what BemiGo says is a first for unmanned logistics vehicles
- Compute stack delivers 550 TOPS, supporting a 7V+1L sensor configuration and 360-degree perception
- Conference appearance positions BemiGo alongside NVIDIA, Amazon, and Google DeepMind as it pursues international commercial deployment
T6 Robot Targets High-Volume Last-Mile Routes
The T6 is positioned for parcel-heavy last-mile and middle-mile delivery applications. Each trip can accommodate 600 to 800 parcels, according to Chen, with a wading depth of 190 mm intended to support operation in adverse weather. The vehicle uses a 7V+1L configuration — seven cameras and one lidar — combined with surround perception coverage. BemiGo said the platform supports automotive-grade safety standards and over-the-air software updates.
The 550 TOPS of onboard compute aligns the T6 with the broader industry shift toward generative AI workloads in autonomous vehicles, where vision-language-action models require significantly more processing headroom than earlier modular ADAS architectures.
SmartVLA Built on NVIDIA Foundation Model
BemiGo’s SmartVLA large model is built on NVIDIA’s Alpamayo 1.5 open-source foundation model and runs on the Jetson Thor compute platform. The company said this represents the first deployment of the model on an unmanned logistics vehicle. The system handles real-world scene understanding and autonomous decision-making functions traditionally split across separate perception, prediction, and planning modules.
End-to-end approaches collapse those layers into a unified neural network, a direction increasingly adopted across the AV industry as developers move away from rules-based driving stacks. BemiGo’s Smartware 2.0 middleware sits beneath SmartVLA and provides what the company describes as a Lego-style modular architecture, allowing hardware and software components to be reconfigured for different vehicle types.
Brain 2.0 Adds Multi-Agent Cloud Orchestration
The Brain 2.0 platform is a cloud-based system that coordinates fleet operations across multiple vehicles and use cases. It pairs a cloud control platform with a closed-loop data system, an AI agent monitoring center, and a mobile management application. Together, the components are intended to handle the full delivery workflow — vehicle dispatch, route management, exception handling, and remote monitoring — within a single multi-agent architecture.
Commercial Context
BemiGo joins a competitive field of Chinese autonomous logistics operators pursuing commercial deployment in last-mile and middle-mile freight. The company said it is working with ecosystem partner Magic Atom on broader embodied intelligence applications and signaled plans to expand commercial services into international markets, though no specific deployment timeline or partner customers were disclosed in the announcement.
Sign up for our weekly emails to stay up to date on the latest news!
