THE CHALLENGE
The construction industry faces a major challenge in adopting robotic automation due to the unpredictable and dynamic nature of job sites, which makes traditional robotic systems, relying on rigid programming or classical planning algorithms, too inflexible for real-world use. While cutting-edge AI technologies like foundation models promise greater adaptability, they are often too expensive and resource-intensive for on-site deployment. Data-driven approaches also require massive proprietary datasets that most construction firms are unwilling or unable to share, limiting their practicality. Teleoperation can fill some gaps but introduces delays and increases the need for human oversight, which reduces efficiency. The result is a costly mismatch between the promise of intelligent robotics and the industry's practical need for affordable, adaptive, and scalable solutions that can operate reliably in messy, changing environments without constant reprogramming or supervision.
OUR SOLUTION
Our solution introduces a cost-effective and scalable AI planning system for construction robots by using lightweight, open-source language and vision-language models instead of expensive commercial alternatives. Built on a modular, Soar-inspired architecture, the system divides planning tasks among multiple cooperating AI agents that interpret real-time sensor data, apply decision rules, and communicate to refine action plans for roles like painting, inspection, and tiling. This approach avoids the need for large proprietary datasets or extensive manual programming, enabling robots to adapt to new tasks with minimal human input. The multi-agent setup delivers strong planning accuracy and flexibility while keeping compute and operational costs low, making it a practical choice for firms looking to automate labor-intensive tasks without sacrificing adaptability or budget.
Figure: Overview of the framework
Advantages:
- Superior task accuracy and temporal planning across multiple construction roles
- Up to 10x lower cost using lightweight open-source models
- Real-time adaptability through direct sensor integration
- Scalable multi-agent architecture for performance-resource flexibility
Potential Application:
- AI-driven multi-agent task planning for construction robots
- Autonomous construction site inspections
- Automated wall painting and floor tiling
