Sidi Smart Driving (03881) has made key progress in mining AI application exploration. Will the world's first mining intelligent “Xiaoyuan” be a mining version of Muse?

AI applications are entering a blowout period with the continuous transition of large model capabilities and a rapid decline in computing power costs. If we talk about the release of consumer Meta Muse, it marks a new stage where individual agents move from “answering questions” to “doing things for others.” So, when AI actually enters the hardest core industrial sites such as mines and ports, who can become that “on-the-job” smart body? Sidi Smart Driving (03881), which is deeply involved in driverless mining cards, was the first to give an answer. On September 23, Sidi Smart Driving released the world's first mining intelligent “Xiaoyuan”.

According to reports, “Xiaoyuan” is the first smart device in the industry to achieve a closed loop from voice input to task execution in a mine production environment. It first landed on the “Yuan Mine” scheduling platform, which can control the entire mine with one sentence. It is worth mentioning that before its official debut, “Xiaoyuan” had already been put into practical use in 15 mines.

The daily operations of the mine have long been plagued by a broad model of “manual operation, information reliance, and decision-making based on experience”. The training cycle for new hires is long, and on-site response is highly dependent on the personal experience of the core dispatcher. This experience-driven model, which is difficult to standardize and replicate, forms an invisible ceiling for mine expansion.

The solution to “Xiaoyuan” is to use a unified intelligent portal to string together the mine's existing digital systems. The dispatcher said that the system automatically completes status inquiries and parameter checks, and generates executable instructions after safety verification. Manual confirmation is required only when high-risk operations are involved.

This system architecture, which can be executed in a closed loop, consists of three layers. The “Yuanshen” mining model is responsible for understanding intentions. The Harness intelligent orchestration engine breaks down complex targets into executable task chains, and heavy loads the world model cloud to perform physical deduction. Unlike general models currently on the market, which can only answer mining knowledge questions, the “Yuanshen” mining model is based on training on mine corpus and process rules, and uses real-time production data as context to complete intention understanding, task reasoning, and plan generation, and can be quickly adapted to different mining sites. The difference between the two is not only the speed of response, but also the depth of understanding of the business and the reliability of the results.

This architecture gives Xiaoyuan the three basic core capabilities — global monitoring, understanding and execution, and advance prediction. It can be on duty 7×24 hours a day to achieve global monitoring and scheduling of all mine equipment and operation scenarios; handle faults and unexpected operating conditions based on site conditions to achieve understandable execution; predict equipment faults, yard congestion, and production trends through historical data and realistic perception, and preempt post-incident response as prior intervention to directly improve production capacity and safety.

The actual data confirms this. Judging from the operational data of the 15 mines that have already deployed “Xiaoyuan”, the three core indicators of safety, efficiency, and cost have all improved significantly. At the safety level, the safety review was passed in one go, the dispatch instructions left marks throughout the process, data was kept on the mining area's internal network, and there were zero changes to safety management standards. In terms of efficiency, batch scheduling was reduced from a few minutes to 3 seconds, production reports were reduced from half a day to automatically generated in 3 minutes, and fault handling efficiency increased by 80%. At the cost level, unit transportation costs were reduced by 5%, bicycle transportation efficiency increased by an average of 11.7%, and 100 unmanned mining cards transported 10,000 square meters of materials every day.

As a new exploration of AI applications in mines by Sidi Smart Driving, “Xiaoyuan” has significantly increased productivity. It has transformed mine scheduling from being highly dependent on personal experience to replicable system capabilities. From the perspective of the development of the entire industry, the entry of a large model into the mine also means that the mine's exploration of AI has entered the deep-water zone. Following this logic, with the large-scale deployment of “Xiaoyuan” in more mines in the future, this set of “replicable system capabilities” is expected to accelerate penetration into the entire industry and push the industry to a new level of intelligence.