Search papers, labs, and topics across Lattice.
This paper introduces a UNet-based neural network for 3D cone localization in autonomous racing, addressing the limitations of traditional computer vision and real-time feasibility of neural networks trained on limited data. The network performs keypoint detection on cones using a custom-labeled dataset, enabling accurate cone position estimation and color prediction. The model demonstrates significant improvements in keypoint accuracy and high-quality end-to-end autonomous system performance.
A UNet trained on a massive custom dataset achieves a significant leap in 3D cone localization accuracy, paving the way for more robust autonomous racing systems.
Accurate cone localization in 3D space is essential in autonomous racing for precise navigation around the track. Approaches that rely on traditional computer vision algorithms are sensitive to environmental variations, and neural networks are often trained on limited data and are infeasible to run in real time. We present a UNet-based neural network for keypoint detection on cones, leveraging the largest custom-labeled dataset we have assembled. Our approach enables accurate cone position estimation and the potential for color prediction. Our model achieves substantial improvements in keypoint accuracy over conventional methods. Furthermore, we leverage our predicted keypoints in the perception pipeline and evaluate the end-to-end autonomous system. Our results show high-quality performance across all metrics, highlighting the effectiveness of this approach and its potential for adoption in competitive autonomous racing systems.
Amazon Science