Agile3D: Adaptive Contention- and Content-Aware 3D Object Detection for Embedded GPUs

Principal Investigator: Somali Chaterji

Efficient 3D perception is critical for autonomous systems---self-driving vehicles, drones---to navigate safely in dynamic environments. Accurate 3D object detection from LiDAR data must handle irregular, high-volume point clouds, variable latency from contention and scene complexity, and tight embedded GPU constraints. Balancing accuracy and latency under dynamic conditions is crucial, yet existing frameworks like Chanakya [NeurIPS '23], LiteReconfig [EuroSys '22], and AdaScale [MLSys '19] struggle with the unique demands of 3D detection. We present Agile3D, the first adaptive 3D system integrating a cross-model Multi-branch Execution Framework (MBEF) and a Contention- and Content-Aware Reinforcement Learning-based controller (CARL). CARL dynamically selects the optimal execution branch using five novel MEF control knobs: encoding format, spatial resolution, spatial encoding, 3D feature extractor, and detection head. CARL uses Direct Preference Optimization (DPO) to finetune branch selection without hand-crafted rewards, presenting the first application of DPO to branch scheduling in 3D detection. Comprehensive evaluations show that Agile3D achieves state-of-the-art performance, maintaining high accuracy across varying hardware contention levels and 100-500 ms latency budgets. On NVIDIA Orin and Xavier GPUs, it consistently leads the Pareto frontier, outperforming existing methods for efficient 3D detection.

Representative Publications

• @article{wang2025agile3d,
   title={Agile3D: Adaptive Contention- and Content-Aware 3D Object Detection for Embedded GPUs},
   author={Wang, Pengcheng and Liu, Zhuoming and Bagchi, Shayok and Xu, Ran and Bagchi, Saurabh and Li, Yin and Chaterji, Somali},
   year={2025}
   publisher={ACM MobiSys}
  }

Keywords: autonomous driving, direct preference optimization, latency-sensitive, LiDAR, resource contention, streaming analytics, streaming videos