Video Friday: Digit Learns to Dance—Virtually Overnight
Digit Learns to Dance—Virtually Overnight
A Breakthrough in Robot Learning
In a remarkable demonstration of the power of artificial intelligence, Digit, a humanoid robot, has learned to dance in a matter of hours. This achievement is a testament to the rapid progress being made in the field of robot learning, where machines are being taught to perform complex tasks with unprecedented speed and accuracy.
The feat was accomplished using a novel approach called sim-to-real reinforcement training, which involves teaching a robot to perform a task in a simulated environment and then transferring that knowledge to the real world. In this case, the training data was obtained from motion capture, animation, and teleoperation methods, allowing Digit to learn new whole-body control capabilities overnight.
GEN-1: A General-Purpose AI Model
The development of GEN-1, a general-purpose AI model, marks a significant milestone in the field of robot learning. This model has been shown to achieve mastery of simple physical tasks with unprecedented speed and accuracy, improving average success rates to 99% on tasks where previous models achieved 64%. GEN-1 also completes tasks roughly 3x faster than state-of-the-art models and requires only 1 hour of robot data for each of these results.
The implications of GEN-1 are far-reaching, as it has the potential to unlock commercial viability across a broad range of applications. While it cannot solve all tasks today, it is a significant step towards the mission of creating generalist intelligence for the physical world.
Open-Source Dataset for Humanoid Robot Whole-Body Teleoperation
Unitree has open-sourced the UnifoLM-WBT-Dataset, a high-quality real-world humanoid robot whole-body teleoperation dataset for open environments. This dataset will continue to receive high-frequency rolling updates and aims to establish the most comprehensive real-world humanoid robot dataset in terms of scenario coverage, task complexity, and manipulation diversity.
MRReP: A Mixed Reality-Based Interface
Autonomous mobile robots operating in human-shared indoor environments often require paths that reflect human spatial intentions, such as avoiding interference with pedestrian flow or maintaining comfortable clearance. MRReP, a Mixed Reality-based interface, enables users to draw a Hand-drawn Reference Path (HRP) directly on the physical floor using hand gestures.
Mirrorbot: Facilitating Serendipitous Interactions
Eye contact, even momentarily between strangers, plays a pivotal role in fostering human connection, promoting happiness, and enhancing belonging. Through autonomous navigation and adaptive mirror control, Mirrorbot facilitates serendipitous, nonverbal interactions by dynamically transitioning reflections from self-focused to mutual recognition, sparking eye contact, shared awareness, and playful engagement.
PAL Robotics' Teleoperation System
PAL Robotics has developed a real-time VR teleoperation setup for its AI-ready mobile manipulator, TIAGo Pro. This system allows precise control of TIAGo Pro's dual arms in Cartesian space, ideal for remote manipulation, AI data collection, and robot learning.
Robust AI's Teleoperation System
Robust AI has developed a teleoperation system that enables users to control robots remotely using a VR interface. This system has the potential to revolutionize the way we interact with robots and has far-reaching implications for industries such as manufacturing, healthcare, and logistics.
iRobot's Home Test Labs
iRobot has developed a series of home test labs where its robots can be tested and trained in a variety of scenarios. This allows the company to develop and refine its robots in a real-world setting, ensuring that they are ready for deployment in a variety of applications.
UR Cobots in Production
UR cobots have been used in production to automate the final "magic 5%" of production, allowing THEMAGIC5 to deliver affordable, custom-fit goggles. This is a testament to the power of automation and the potential for cobots to revolutionize manufacturing.
Sanctuary AI's Hydraulic Hands
Sanctuary AI has developed a series of hydraulic hands that can be used for a variety of tasks, including manipulation and grasping. The company's proprietary hydraulic hand has been shown to autonomously manipulate a lettered cube, continuously reorienting it to match a specified goal.
Yuxing 3-06: A Space Refueling Station
China's Yuxing 3-06 commercial experimental satellite has recently completed an in-orbit refueling test and verification of key technologies. This marks a significant milestone in the development of space refueling and has far-reaching implications for the future of space exploration.
Tokyo Robotics' Humanoid Robot
Tokyo Robotics has developed a humanoid robot that can perform a variety of tasks, including natural walking, whole-body teleoperation, and motion tracking. The control policies for this robot were trained using large-scale parallel reinforcement learning (RL).
Serendix's 3D Printing Technology
Serendix has developed a 3D printing technology that can be used to create complex structures, including buildings. The company's technology was used to replace an old wooden building at Japan Railway West's Hatsushima railway station.
Humanoid, SAP, and Martur Fompak's Joint Proof of Concept
Humanoid, SAP, and Martur Fompak have teamed up to test humanoid robots in automotive manufacturing logistics. This joint proof of concept explores how robots can streamline operations, improve efficiency, and shape the future of smart factories.
MIT Robotics Seminar: Avian Inspired Drones
Dario Floreano from EPFL presented a talk on "Avian Inspired Drones" at the MIT Robotics Seminar. This talk explored the use of avian-inspired drones for a variety of tasks, including surveillance and search and rescue.
MIT Robotics Seminar: Good Old-Fashioned Engineering Can Close the 100,000 Year 'Data Gap' in Robotics
Ken Goldberg from UC Berkeley presented a talk on "Good Old-Fashioned Engineering Can Close the 100,000 Year 'Data Gap' in Robotics" at the MIT Robotics Seminar. This talk explored the use of traditional engineering techniques to address the data gap in robotics.
Conclusion
The field of robot learning is rapidly advancing, with machines being taught to perform complex tasks with unprecedented speed and accuracy. The development of GEN-1, a general-purpose AI model, marks a significant milestone in this field and has the potential to unlock commercial viability across a broad range of applications. As the field continues to evolve, we can expect to see even more impressive achievements in the years to come.
