Industrial IoT

How to Use Language Models to Advance Robotics: Insights from Recent Research Papers

The process industry and machinery sector have always been at the forefront of adopting innovative technologies. Two recent research papers propose approaches to using large language models (LLMs) in an industrial context that might inspire professionals working in this field. Eureka and GenSim, two advanced AI systems, shed light on LLMs' potential capabilities, and the implications for industrial companies are potentially profound. Eureka uses LLM's to motor control tasks and GenSim creates rich simulation environments for robotic machinery.

Eureka: LLM-powered reward design algorithm for motor control tasks

Eureka, another groundbreaking AI system, has been designed to learn complex low-level motor control tasks based on high-level reasoning. The system has two layers; LLM instructs a learnable neural network. The outer layer uses GPT-4 to write and refine code for a reward function. The reward function is then used to train a reinforcement learning algorithm to conduct a given task.

For industrial companies, this means that robots can be trained to perform complex tasks with a level of dexterity previously thought impossible. Eureka generates reward functions that outperform expert human-engineered ones on 83% of the tasks (n=29) by an average improvement margin of 52%.

The Eureka model also enables a new gradient-free in-context learning approach to reinforcement learning from human feedback (RLHF), readily incorporating human inputs to improve the quality and safety of the generated rewards without model updating.

GenSim: A New Era of Robotic Policies

GenSim, as explained in the research paper, leverages LLMs to generate rich simulations. This addresses the scarcity of high-quality data for robot task training.

The GenSim approach has two modes: goal-directed generation and exploratory generation. The former involves giving a target task to the LLM, which then proposes a task curriculum to solve the target task. The latter allows the LLM to bootstrap from previous tasks and propose novel tasks that would be helpful in solving more complex tasks.

With GenSim, industrial companies can train their robotic policies on various tasks, enhancing their versatility and adaptability. This means that robots can be more effectively used in a wider range of applications, improving efficiency and productivity.

Implications for the Process Industry and Machinery

The GenSim and Eureka models have far-reaching implications for the process industry and machinery. Here are some key takeaways:

  1. Enhanced Robotic Capabilities: With these models, robots can be trained to perform a wider range of tasks with greater efficiency and precision. This can significantly boost productivity and reduce the risk of human error.
  2. Improved Safety: The ability to incorporate human feedback in real-time means that safety measures can be more effectively implemented, reducing the risk of accidents and ensuring a safer working environment.
  3. Cost Savings: By training complex tasks in a simulation environment, companies can save on costs and reduce the time taken to complete tasks. 
  4. Increased Versatility: The ability to train robots on a wide array of tasks means that they can be more versatile and adaptable, able to handle a wider range of tasks and adapt to changing circumstances.


The GenSim and Eureka models represent a significant leap forward in using LLMs to perform tasks with robots. For industrial companies in the process industry and machinery, these systems could offer a way to enhance productivity, improve safety, and save costs. As these technologies evolve, we can expect to see even more exciting developments. 

At Brightly, we not only understand the transformative potential of these advanced AI models but also the need to future-proof the current technical infrastructure. With our extensive experience in delivering end-to-end IIoT solutions, we are uniquely positioned to help industrial and machinery companies navigate this rapidly evolving landscape. We believe in the power of technology to drive efficiency, safety, and productivity. As the market continues to evolve, Brightly is committed to leading the way, helping our clients harness the power of data, AI, and machinery to optimize their operations and stay at the forefront of their industries. With Brightly as your partner, you can confidently step into the future of industrial automation.


Joanna Purosto

Entrepreneurial-minded and nerdy marketing professional. Joanna has years of experience building marketing functions for start-ups and scaleups in the AI and IT sectors. She has an M.Sc. in Economics, majoring in Entrepreneurship and minoring in Artificial Intelligence.