Summary

  • ECOPEACE’s “ECOBOT,” an autonomous water pollutant detection and purification solution based on AI vision recognition.
  • Deployment of Rebellions NPU servers within UAE local infrastructure to support ECOBOT operations.
  • Demonstrated operational efficiency in a Physical AI environment, achieving a 2x improvement in performance-per-watt compared to GPUs.

Challenge

In the Middle East, where the oil and gas industries are highly developed, managing wastewater and oil film generated during crude oil production is a critical requirement. Manual treatment of aquatic environments such as reservoirs is time-consuming. Furthermore, if pollutants are left untreated for extended periods, they degrade, necessitating additional processing and resulting in long-term economic losses. To address these issues, the UAE focused on ECOPEACE’s AI vision-based pollutant detection and autonomous purification solution.

Solution

ECOPEACE implemented an intelligent water purification robot system utilizing a central server powered by the Rebellions ATOM™ NPU. When the ECOBOTs operating at Dubai’s Al Jadaf Port transmit captured video data, the NPU server located in Abu Dhabi performs precise identification via object detection models while simultaneously calculating 3D distances through depth estimation models. Upon receiving these results, the robots approach the pollutants and activate oil recovery pumps to perform suction and filtration via internal oil-water separation components and filter designs.

High-precision AI inference based on high-resolution data, which is difficult to process on edge devices, is handled by the central server. This architecture reduces the system load on the robots and maximizes battery efficiency, supporting sustainable operations. The NPU server consumes less power than GPU-based servers and offers lower deployment and operational costs, enabling the establishment of an economical AI infrastructure.

Result

Through the adoption of the Rebellions ATOM™ NPU, ECOPEACE officially verified via TTA (Telecommunications Technology Association) certification that it is possible to achieve twice the performance-per-watt compared to GPUs. Based on these results, the NPU-based inference infrastructure was successfully established in the UAE, marking a significant precedent for the global expansion of Physical AI solutions centered on Korean AI semiconductors. This case demonstrates that the Rebellions NPU is a proven alternative capable of replacing GPUs in the Physical AI domain, where multiple AI models must interact, moving beyond simple standalone inference tasks.

Appendix: NPU User Guide

Get Started: Prepare pre-trained model → Compile model using RBLN compiler → Load compiled model and run inference