Hitachi is teaching AI how to dig

Hitachi Construction Machinery will begin research and development in October 2026 on Physical AI trained to replicate the skills of human hydraulic excavator operators, aiming to eventually achieve fully autonomous operation on construction, mining and disaster sites.

The push is driven by labour shortages across the construction and mining sectors, alongside a growing need to improve safety and productivity in harsh working environments. Hitachi also points to the rising frequency of severe natural disasters and the importance of maintaining a stable supply of mineral resources as factors behind the project.

The project has been selected for GENIAC (Generative AI Accelerator Challenge), a national initiative run by Japan's Ministry of Economy, Trade and Industry (METI) and the New Energy and Industrial Technology Development Organization (NEDO) to strengthen the country's generative AI capabilities. Hitachi will develop the technology in partnership with Jizai, Inc. and the Nara Institute of Science and Technology (NAIST).

The AI will be trained on real operator lever inputs, camera footage and work instructions, learning to read site conditions — including terrain, weather and machine status — and determine the appropriate excavator movements in response.

Hitachi frames Physical AI around a three-stage loop: perception, decision and action. First, the system builds an understanding of its surroundings by reading the shape and position of objects, the terrain and prevailing conditions. From there, it moves into decision-making, working out the optimal sequence of tasks needed to complete the job safely and efficiently. Finally, it executes those decisions as physical lever operations on the machine itself.

Hitachi draws a clear distinction between this and conventional automation, which follows fixed, pre-programmed motions. Physical AI, by contrast, is meant to genuinely interpret its environment and adapt its actions accordingly, task by task.

To build the training dataset, Hitachi will collect extensive footage and operation logs from human operators, deliberately capturing not just routine work but also error recovery, and a wide range of weather, environmental and machinery conditions.

The three-way collaboration divides responsibilities as follows: Hitachi will coordinate the project and collect excavator operational data; Jizai will research the data utilisation and training environment; and NAIST will lead development of the underlying Physical AI foundation model.

Hitachi says the long-term goal is to deploy the technology across disaster response, material handling, forestry, mining, demolition and civil engineering.

The announcement comes as Hitachi Construction Machinery prepares to rebrand as LANDCROS Corporation from April 2027.