Shanghai-based robotics company Fourier is at the forefront of enhancing humanoid robots for real-world applications, utilizing NVIDIA’s cutting-edge technology. The company recently expanded its GRx humanoid robot series with the introduction of GR-2, which offers significant advancements in hardware design, adaptability, and dexterity, according to NVIDIA’s blog.
Developing Humanoid Robot GR-2 with NVIDIA Isaac Gym
Fourier’s development of the GR-2 humanoid robot leverages NVIDIA Isaac Gym, a platform for reinforcement learning, to streamline the training process. This approach allows for the simulation of complex, real-world scenarios, thereby reducing testing time and costs. The company is transitioning its workflows to NVIDIA Isaac Lab, which offers an open-source modular framework aimed at simplifying robot learning.
The use of NVIDIA’s tools enables Fourier to simulate intricate multi-robot scenarios and varied environments, leading to improved AI decision-making. This simulation includes pretraining grasping algorithms, which significantly reduces real-world trial and error, saving both time and resources.
Optimizing AI for Real-World Robotics
Fourier has optimized the GR-2’s AI capabilities using NVIDIA TensorRT for real-time inference optimization and CUDA libraries for enhanced processing. This allows the robots to perform complex maneuvers such as the floor-to-stand transition with an 89% success rate after 3,000 iterations, a substantial improvement over traditional methods.
The integration of these technologies not only accelerates the training process but also enhances the robots’ real-time motion control and AI-driven decision-making, setting new standards for human-robot interaction in various industries, including healthcare and scientific research.
Exploring Next-Generation Robotic Capabilities
Fourier’s adoption of NVIDIA technologies has reduced training times and improved simulation accuracy, facilitating better collaboration between the company’s engineering and R&D teams. This has opened new possibilities for complex AI functions, such as language models and predictive analytics, which were previously too resource-intensive to implement.
Fourier CEO Alex Gu highlighted these advancements, stating, “The improvements in real-time motion control and AI decision-making are setting new benchmarks for humanoid robotics, particularly in sectors like service, academic research, and medical rehabilitation.”
For more detailed information, visit the NVIDIA blog.
Image source: Shutterstock