ABB Robotics Crowns AI Startup Challenge 2024 Winners

T-Robotics and Mbodi were named winners of ABB Robotics’ AI Startup Challenge 2024, highlighting a pivotal step in generative AI solutions for industrial robotics. The winners were selected from 100 hopeful applicants and received $30,000 in funding. ABB will work closely with T-Robotics and Mbodi to support their projects and launch their first commercial applications in 2025. The two winners will also have access to ABB’s startup accelerator, SynerLeap, to facilitate continued collaboration and development.

The world of industrial robotics will receive a generative AI boost from ABB AI Startup Challenge winners Mbodi and T-Robotics. Image used courtesy of ABB Robotics

Generative AI in Robotics

Generative AI is a branch of artificial intelligence based on how the human brain takes in, processes, and generates information. ChatGTP is a large language model based on the deep learning method. More specifically, it is termed a transformer, a type of neural network that can focus on key elements within a text to mine text data and pick out the most suitable response.

Manufacturers can collate data, analyze it, and generate solutions and insights to improve operational efficiencies by applying generative AI to industrial processes. Generative AI, including approaches such as simulation-to-real transfer (Sim2Real), is transforming how industrial robots learn new skills and adapt to different environments. Robots initially learn in virtual environments using simulators such as NVIDIA Isaac Sim, while generative models such as generative adversarial networks (GANs) or diffusion models generate realistic, synthetic sensor data. This method simulates various settings, allowing robots to expand their abilities and respond appropriately to real-world industrial scenarios. For instance, a robot arm trained to arrange objects in a simulation may adapt to real-world settings with little retraining, which cuts costs and improves flexibility.

The ABB AI Startup Challenge 2024 heralds Mbodi and T-Robotics as the winners among 100 applicants. Video used courtesy of ABB Robotics

Another use of generative AI in robotics is data augmentation. Restricted real-world datasets are expanded using generative models such as diffusion models or variational autoencoders (VAEs) which provide artificial training data. This allows robots to learn complicated tasks (through reinforcement learning), such as grabbing or carefully piecing together many components with greater accuracy and adherence to quality standards.

Generative AI speeds up training and increases robot efficiency and adaptability by eliminating the need for time-consuming manual programming, which better prepares robots to manage changing task requirements.

T-Robotics' AI-powered solution will enable complex robot programming via the use of natural language. Video used courtesy of T-Robotics

T-Robotics

T-Robotics has developed ActGTP, a platform that enables customers to expedite robot programming for machine tending applications. The idea is similar to ChatGPT: users input text prompts, which are then combined with AI-based problem-solving, embedded known skills, and visual and other data (such as haptics) to generate programs for machine-tending robots.

ActGPT is designed to be a flexible platform that works with a variety of commercial robots. T-Robotics is cooperating with ABB to launch its first commercial solution in 2025.

Mbodi’s AI-powered solution is designed to teach robots new tasks in a matter of minutes. Video used courtesy of Mbodi

Mbodi

Mbodi aims to improve manufacturing with a user-friendly, flexible, and practical AI-powered platform. The company's platform utilizes generative AI to help robots learn, adapt, and overcome real-world industrial challenges. The company's technology enables robots to learn new abilities quickly, adapt to changing environments, and perform natural language directives in as little as 0.5 seconds, streamlining conventional robotics programming and boosting the efficiency of operations.

Mbodi's platform combines the flexibility of generative AI with well-established robotics technologies to develop modular, configurable systems that can adapt to a variety of hardware and application cases. Mbodi also promotes open-source innovation by providing tools such as its embodied-agents library, which incorporates advanced AI models into robotics applications. Mbodi's programs aim to create self-learning systems that tackle tasks in high-mix, low-volume manufacturing environments.