Kajima Begins Full-Scale Rollout of Robots with FieldAI Autonomy

Kajima announced on November 5, 2026 that it had begun full-scale deployment of robots using FieldAI’s physical AI autonomy software on construction sites, with more than 2,000 autonomous missions completed to date and deployments at more than 10 sites in Japan in roughly a year under an enterprise agreement.

The missions have spanned different robot types, tasks and sites, and the robots are in continuous use on active projects where structures and the placement of materials and equipment change daily. That operating record moves the initiative beyond limited demonstrations. Kajima and FieldAI have collaborated since 2023, refining the technology on real job sites, and an enterprise-level service agreement reached in 2025 shifted the work from individual projects to full-scale deployment.

Kajima is using the deployment to reduce routine construction-management workload as it responds to construction labor shortages. Engineers’ work spans site rounds, visual documentation and 3D scanning alongside the technical judgments required for safety, quality and schedule management. Assigning some of the routine work to robots is intended to give engineers more capacity for advanced judgment and safety management.

A General-Purpose Autonomy Unit

The system is built around an autonomy unit that can be mounted on different robots. Cameras capture surroundings such as obstacles and no-entry areas, while LiDAR uses reflected laser light to measure distance and shape and collect point-cloud data for mapping. An edge AI device processes that imagery and point-cloud information onboard rather than in the cloud; Kajima said it determines a safe travel direction in real time.

That local processing supports autonomous operation indoors, in underground locations where satellite positioning is unavailable, and in places where external communications are difficult. The onboard software uses FieldAI’s Field Foundation Models, which Kajima said combine data-driven learning, physics-based reasoning and uncertainty quantification so the system can handle previously unlearned situations and changing environments while continuing to operate safely. Unlike conventional robot control technology that requires customization for specific sites, the software is designed to serve as a general-purpose autonomy layer across different robots and tasks.

Autonomous Rounds Create Site Records

On first startup, a robot explores an unknown area on its own and generates a three-dimensional map. During later rounds, it avoids obstacles, captures photographs and point-cloud measurements, and updates the map as conditions change. After finishing its work, the robot returns to its charging station and uploads the collected data, which is automatically organized and made available on the web.

Users can also select a destination from a tablet or PC, sending the robot there autonomously and viewing real-time video from that location. For progress management, autonomous rounds collect 360-degree photographs and high-resolution color point-cloud data. The resulting records can be used to determine construction status and dimensions, streamlining progress checks and giving teams a recurring digital record of conditions across the site.

Security Patrols and Emergency Response

Security patrols can run on non-working days, late at night or early in the morning. When a robot detects an anomaly, it flashes high-intensity LEDs and issues an audio alert, while responsible personnel can view the captured video in real time. Kajima said this supports earlier anomaly detection and a rapid initial response.

For disaster recovery after earthquakes, heavy rain and other emergencies, Kajima said the robots can enter before site safety has been confirmed and collect photographs, point-cloud data and live video in place of people. That is intended to accelerate the initial response without sending personnel into areas that have not yet been assessed.

Kajima plans to add job sites and applications progressively and to apply the autonomy software to other technologies under development, with the stated goal of extending autonomy across machinery and improving safety and productivity as personnel and machines share site work.

The deployment establishes a practical division of labor: autonomous systems repeatedly capture and organize site information, while engineers concentrate on technical judgment, safety, quality and schedules. The planned expansion into other machinery raises the central question of how widely one shared autonomy layer can support that model across Kajima’s construction operations.