In the fight against COVID-19, many robots replace human employees in various tasks that involve a risk of infection. Among these tasks, the fundamental problem of navigating robots among crowds, named robot crowd navigation, remains open and challenging. Therefore, we propose HGAT-DRL, a heterogeneous GAT-based deep reinforcement learning algorithm. This algorithm encodes the constrained human-robot-coexisting environment in a heterogeneous graph consisting of four types of nodes. It also constructs an interactive agent-level representation for objects surrounding the robot, and incorporates the kinodynamic constraints from the non-holonomic motion model into the deep reinforcement learning (DRL) framework. Simulation results show that our proposed algorithm achieves a success rate of 92%, at least 6% higher than four baseline algorithms. Furthermore, the hardware experiment on a Fetch robot demonstrates our algorithm’s successful and convenient migration to real robots.

The advancements in hardware have drawn more attention than ever to high-quality offline rendering with modern stream processors, both in the industry and in research fields. However, the graphics APIs are fragmented and existing shading languages lack high-level constructs such as polymorphism, which adds complexity to developing and maintaining cross-platform high-performance renderers. We present LuisaRender 1 , a high-performance rendering framework for modern stream-architecture hardware. Our main contribution is an expressive C++-embedded DSL for kernel programming with JIT code generation and compilation. We also implement a unified runtime layer with resource wrappers and an optimized Monte Carlo renderer. Experiments on test scenes show that LuisaRender achieves much higher performance than existing research renderers on modern graphics hardware, e.g., 5--11× faster than PBRT-v4 and 4--16× faster than Mitsuba 3.

Spherical Harmonics (SH) have been proven as a powerful tool for rendering, especially in real-time applications such as Precomputed Radiance Transfer (PRT). Spherical harmonics are orthonormal basis functions and are efficient in computing dot products. However, computations of triple product and multiple product operations are often the bottlenecks that prevent moderately high-frequency use of spherical harmonics. Specifically state-of-the-art methods for accurate SH triple products of order n have a time complexity of O ( n 5 ), which is a heavy burden for most real-time applications. Even worse, a brute-force way to compute k -multiple products would take O ( n 2 k ) time. In this paper, we propose a fast and accurate method for spherical harmonics triple products with the time complexity of only O ( n 3 ), and further extend it for computing k -multiple products with the time complexity of O ( kn 3 + k 2 n 2 log ( kn )). Our key insight is to conduct the triple and multiple products in the Fourier space, in which the multiplications can be performed much more efficiently. To our knowledge, our method is theoretically the fastest for accurate spherical harmonics triple and multiple products. And in practice, we demonstrate the efficiency of our method in rendering applications including mid-frequency relighting and shadow fields.

Deep learning techniques such as convolutional neural network and capsule network have attained good results in iris recognition. However, due to the influence of eyelashes, skin, and background noises, the model often needs many iterations to retrieve informative iris patterns. Also because of some nonideal situations, such as reflection of glasses and facula on the eyeball, it is hard to detect the boundary of pupil and iris perfectly. Under such a circumstance, discarding the rest parts beyond the boundary may cause losing useful information. Hence, we use Gaussian, triangular fuzzy average, and triangular fuzzy median smoothing filters to preprocess the image by fuzzifying the region beyond the boundary to improve the signal-to-noise ratios. We applied the enhanced images through fuzzy operations to train deep learning methods, which speeds up the process of convergence and also increases the recognition accuracy rate. The saliency maps show that fuzzified image filters make the images more informative for deep learning. The proposed fuzzy operation of images may be a robust technique in many other deep-learning applications of image processing, analysis, and prediction.