Banma Intelligent Releases AutoOmni 2.0 End-Side Model, Debuts AutoClaw 2.0 Cockpit Solution

Banma Intelligent co-CEO Hao Fei said the core value of edge intelligence is opening a new interaction paradigm. He said whole-car, all-time wake-free operation, along with context and memory based on the end-side model, are key elements of that paradigm, making proactive services that better understand the user possible.

At Banma Intelligent's Yunqi Conference booth, the smart-driving systems on the Denza Z9GT, IM LS6, Shenxingzhe 8 and Hongqi Tiangong 06 were God's Eye, Momenta, Huawei Qiankun and Zhuoyu, respectively, but all four vehicles chose Banma Intelligent's YuanShen AI for their smart cockpits. YuanShen AI has achieved service coverage for 68% of mainstream automakers, according to the company.

Banma Intelligent chief product officer Cai Ming said AI being installed in a car does not mean winning users, and deep use is the real battlefield. He said the premise for AI to conduct complex planning and then effectively organize capabilities from different parties is understanding the user. AI must understand the 'surface world' made up of physical space, rules and trajectories, as well as the 'inner world' made up of personal intent, preferences and life scenarios, then keep evolving in a loop of perception, understanding, execution and feedback, Cai said.

According to Cai, YuanShen AI is moving from cockpit AI toward Personal AI and has formed 'three connections': connecting cloud and vehicle compute and services; connecting multiple user terminals and opening up all-time, all-scenario data; and connecting users and services to integrate services across apps and terminals.

Xiao Ruizhe, founder and CEO of AI data terminal company YoooClaw, said that with Banma Intelligent, and under the premise of data compliance, the company is connecting users' 'context' during one hour in the car and 23 hours outside it, so that the value of data and intelligence can be brought into play. At the Banma Intelligent booth, the new AutoClaw 2.0 real-vehicle solution and a new ecosystem agent real-vehicle solution built with YoooClaw have put the 'three connections' initially on the road, according to QbitAI.

During a forum, Banma Intelligent officially released AutoOmni 2.0-23B-A3B. The model uses a MoE mixture-of-experts architecture. In smart-cockpit scenarios, its ordinary task capability is comparable to a cloud model with 10 times the parameter scale, while its complex task processing capability is equivalent to 80% to 90% of such a cloud model, the company said. The Yunqi Conference site showed a real-vehicle solution using the model and an end-cloud integrated Agent architecture. For fuzzy intent and multi-task complex navigation, vehicle control, multimedia and cross-domain tasks, AutoOmni 2.0 can execute with low latency, privacy protection and stable, smooth operation, according to the company.

AutoOmni is described by Banma Intelligent as the industry's first full-modal end-side large model. In one year, among companies that put end-side models into cars, 63% chose Banma Intelligent, the company said.

Banma Intelligent chief technology officer Si Luo said, 'Engrave the end model on the chip, and engrave safety in the heart.' He said AutoOmni has cooperated with 10 chip companies at home and abroad, and that software-hardware coordination supports stable concurrency of more than eight multi-tasks, accelerates large-model inference by five to six times, achieves quantization fidelity of more than 99% and reduces runtime memory use by more than 50%. The automotive-grade end-side AI security defense-in-depth system does more than privacy protection and data compliance; it also balances security and performance overhead, Si Luo said.

Si Luo demonstrated end-side intelligence scenarios including whole-car, all-time wake-free operation, proactive intelligent welcome and contactless smart parking payment. Starting with dozens of proactive intelligence scenarios, AutoOmni opens an AI-native new paradigm for proactive intelligence in cars, according to the company.

'Self-driving cars are the largest pilot testing ground and scaled stress-testing platform for embodied intelligence,' Si Luo said. He said Banma Intelligent is evolving from AutoOmni toward EdgeOmni, expanding around users from single-agent intelligence to swarm intelligence, so that AI terminals better understand 'my world.'

QbitAI reported that the article was provided by Banma Intelligent and reposted with authorization, and that the views belong to the original author.