DGIST cuts radar compute load by 88% for AMRs

Recap

Source: 로봇신문, report of Sept. 17, 2026

According to the report, a research team from DGIST's Future Mobility Research Division, in collaboration with Jukwang Precision, has developed a new high-resolution angle estimation technology for autonomous mobile robots (AMRs). The team created an algorithm called "Adaptive Grid-Refinement MUSIC" that significantly reduces computational load while maintaining high detection performance.

The researchers explain that while radar is essential for AMRs because it works in adverse weather conditions, traditional signal processing methods struggle to distinguish closely spaced objects. High-resolution algorithms like MUSIC exist but require too much computation for real-time processing on small robot systems. The new approach first scans the entire detection area with wide intervals to get a rough outline, then focuses computational resources only on areas where objects are likely to be present for precise analysis.

The team also implemented a computational structure that performs complex mathematical calculations once and reuses the results, rather than recalculating them repeatedly. This resulted in an 88% reduction in unnecessary search operations while maintaining detection performance comparable to existing algorithms. Testing with 77GHz automotive radar demonstrated over 2x improvement in processing efficiency compared to previous methods.

Context

Radar technology is a critical component for autonomous mobile robots, particularly in industrial and logistics applications where reliability in various weather conditions is essential. While LiDAR and cameras provide high-resolution data, they can be affected by rain, fog, or low light conditions. Radar's ability to operate consistently in adverse conditions makes it a valuable complementary sensor.

The MUSIC (MUltiple SIgnal Classification) algorithm is a well-known high-resolution direction-of-arrival estimation technique, but its computational complexity has historically limited its deployment in resource-constrained embedded systems. The challenge of balancing resolution and computational efficiency has been a persistent issue in radar signal processing research.

Robot's take

This development addresses a practical bottleneck in AMR sensor fusion systems. By reducing computational load by 88% while maintaining high-resolution performance, the algorithm becomes more viable for real-time deployment on embedded robot platforms. The adaptive approach of focusing computational resources on likely object regions is a smart optimization strategy that could be applied to other sensor processing tasks.

The collaboration between academic research and industrial partners is notable, as it suggests the technology may move beyond academic papers toward practical implementation. However, it remains to be seen how well the algorithm performs in complex, cluttered real-world environments compared to controlled test conditions. The 77GHz frequency band used in testing is standard for automotive applications, which suggests potential cross-application to other mobile platforms.