No abstract

Hybrid machine learning algorithms that combine deep learning with probabilistic inference techniques provide highly accurate scene perception for robot manipulation. In particular, a 2-stage approach that combines object detection using convolutional neural networks with Monte-Carlo sampling for pose estimation has been shown to perform particularly well under adversarial scenarios. Unfortunately, this accuracy comes at the cost of high computational complexity, which affects runtime, resource utilization, and energy consumption. This paper describes various challenges in developing complexity-aware techniques for robust robot perception and presents a novel hardware accelerator that addresses these challenge. Experimental results show our design is at least 30% faster and consumes 97% less energy compared to an implementation on a high-end GPU. Compared to a low-power GPU implementation, our design is 95% faster while consuming 96% less energy, demonstrating that accurate, energy-efficient scene perception is possible in real time with targeted hardware acceleration. CCS CONCEPTS• Hardware → Hardware accelerators; • Computing methodologies → Rasterization; • Computer systems organization → Real-time system architecture.

Algorithms based on Monte-Carlo sampling have been widely adapted in robotics and other areas of engineering due to their performance robustness. However, these samplingbased approaches have high computational requirements, making them unsuitable for real-time applications with tight energy constraints. In this paper, we investigate 6 degree-of-freedom (6DoF) pose estimation for robot manipulation using this method, which uses rendering combined with sequential Monte-Carlo sampling. While potentially very accurate, the significant computational complexity of the algorithm makes it less attractive for mobile robots, where runtime and energy consumption are tightly constrained. To address these challenges, we develop a novel hardware implementation of Monte-Carlo sampling on an FPGA with lower computational complexity and memory usage, while achieving high parallelism and modularization. Our results show 12X-21X improvements in energy efficiency over low-power and high-end GPU implementations, respectively. Moreover, we achieve real time performance without compromising accuracy.