All articles tagged with #deep reinforcement learning
Disney’s Imagineering, using NVIDIA’s Kamino simulator and deep reinforcement learning, trained a self-walking Olaf animatronic to balance and move on a moving boat for the World of Frozen at Disneyland Paris. Olaf demonstrated onstage at NVIDIA’s GTC keynote, illustrating how GPU-based simulation speeds character development and real-world navigation for park experiences.
Nuclear fusion experiments in tokamaks have achieved significant success, but plasma disruption due to tearing instability remains a critical issue. To address this, researchers have developed an AI-based tearing-avoidance system using deep reinforcement learning (RL) to actively control the beam power and plasma shape in real time, maintaining low tearability and high plasma pressure. Experiments at the DIII-D tokamak demonstrated the system's ability to avoid disruptive tearing instability, showcasing the potential for AI-accelerated real-time instability-avoidance techniques in fusion energy research.
MIT researchers have developed an AI-based method to improve safety and stability in autonomous robots, successfully addressing the ‘stabilize-avoid’ problem. Using a two-step approach involving deep reinforcement learning and mathematical optimization, the method was effectively tested on a simulated jet aircraft. This could have future applications in dynamic robots requiring safety and stability, like autonomous delivery drones.