Autonomous systems are revolutionizing our lives in a big way, but taking them from the lab to the real world has always been a challenge, with computational demands to the gap between simulations and reality being the main constraints. However, the team at Technology Innovation Institute (TII) and New York University (NYU) has come up with a groundbreaking solution, which promises to transform the world of autonomous intelligence.
TII and New York University’s Agile Robotics and Perception Lab have unveiled RLtools, a reinforcement learning library poised to tackle these obstacles and push the field forward. And the main highlight is that it’s open source and available to all.
RLtools is the result of intensive research and development efforts under the joint project, ‘Learning to Fly in Seconds’, a significant milestone in autonomous systems research. In a first, high-speed, end-to-end drone controllers have been trained on a standard commercial-grade computer, ushering in a new era of efficiency and accessibility in training AI models.
What’s the challenge?
The challenges that have traditionally hindered the seamless integration of autonomous systems are diverse, including the computational burden of training AI models, disparity between simulated and real-world environments, and compatibility issues with existing deep learning frameworks.
Here’s the solution
RLtools addresses these problems, boasting a remarkable 75x speed-up compared to popular libraries, thereby significantly reducing training time and resource requirements.
Moreover, RLtools excels in resource efficiency, enabling training on standard laptops or even directly on microcontrollers. This accomplishment represents a paradigm shift in the deployment of autonomous systems, democratizing access to deep reinforcement learning to an unprecedented level.
The real-time performance of RLtools-trained controllers surpasses state-of-the-art controllers used on drones globally. Furthermore, RLtools addresses deployment challenges by being directly deployable on microcontrollers, marking the first-ever training of a deep reinforcement learning algorithm on such devices.
The collaboration on RLtools underscores the exceptional research capabilities of both TII and NYU, reflecting their commitment to innovation and adaptability. Looking ahead, the two entities will continue to enhance RLtools, expanding their suite of algorithms to enhance accuracy and compatibility across platforms.
What’s the next big thing?
The ultimate goal is to develop a single, all-encompassing controller capable of autonomous operations and real-time learning. This integrated approach will establish a unified and resilient system, capable of navigating diverse environments with maximum precision and efficiency.
RLtools represents a major move in the realm of autonomous intelligence. With its groundbreaking capabilities and collaborative spirit, it paves the way for a future where autonomous systems seamlessly integrate into our daily lives, transforming industries and enriching human lives.
Click here to find out more about RLtools: https://github.com/rl-tools/rl-tools