Post by Jaime Andres Rincon Arango

While the other model continues training on the physical robot 🤖🦾, on my second workstation we are making progress on the development of a Gaussian Splatting-based simulation system 🌍✨ for creating highly realistic environments. To generate the Gaussian Splatting map, we are using #Marble Pro from World Labs, a platform focused on creating navigable and spatially consistent 3D worlds from images, videos, and other multimodal data 📷🌐. The goal is to explore possible solutions to one of the major challenges in Physical AI: the large-scale and efficient collection of training data for intelligent models 📷🧠. As always, the biggest challenge remains computational power 💻⚡. Real-time rendering, inverse kinematics (IK), physics simulation, and, in the near future, the integration of ACT models all add significant computational demands. Once I manage to integrate the entire ecosystem together... I hope the result will be amazing! 🚀🤯 At this point, I have come to a very scientific conclusion: what I really need is funding 💸😂. GPUs, storage, simulation, training, more GPUs... and probably even more GPUs 😅🔥. Of course, this approach is not entirely new, and several companies are already using similar technologies to generate digital worlds and synthetic datasets. However, in our case, the main objective is educational 🎓. We want to bring these cutting-edge technologies closer to our fourth-year students so they can experiment with the latest tools in robotics, computer vision, simulation, and machine learning. The idea is not only for students to use these technologies, but also to understand how to build, adapt, and apply them to real-world problems in the next generation of intelligent systems 🚀🧠. #PhysicalAI #GaussianSplatting #WorldLabs #MarbleAI #Robotics #ArtificialIntelligence #Simulation #DigitalTwins #ComputerVision #MachineLearning #EmbodiedAI #Education #EngineeringEducation #ACTLab #RobotLearning #ACT

Video Content