Boston Dynamics, with support from Hyundai, is advancing the development of its Atlas humanoid robot. The latest iteration features an all-electric design, an AI system powered by Nvidia microchips, and significantly expanded mobility. Training methods include supervised learning, motion capture, and extensive simulations, enabling Atlas to perform a range of complex physical tasks. While demonstrating advanced capabilities, the robot still faces challenges in mastering routine daily tasks and navigating varied environments, as the company works toward future integration into industrial and potentially broader applications.
Atlas Robot: A New Generation of Humanoid Development
Boston Dynamics, a Massachusetts-based company, is at the forefront of humanoid robot development. An 88% stake in the company is held by the South Korean carmaker Hyundai, providing significant support for its ambitious projects. The company is currently testing a new generation of its humanoid robot, Atlas.
In October, a 5-foot-9-inch, 200-pound Atlas unit was put through its paces at Hyundai's new Georgia factory. Here, it practiced autonomously sorting roof racks for the assembly line, showcasing its potential for industrial applications. Robert Playter, CEO of Boston Dynamics, who leads the company's humanoid development, has highlighted the growing industry interest in general-purpose robots.
Technological Evolution and Advanced Design
The Atlas robot has undergone a significant transformation. From a bulkier, hydraulic model showcased in 2021, it has evolved into a sleeker, all-electric design that incorporates an AI system powered by Nvidia microchips. This marks a pivotal shift from reliance on engineer-written algorithms to machine learning-based approaches for generating movement. Scott Kuindersma, head of robotics research at Boston Dynamics, attributes the robot's enhanced learning capabilities to a focused strategy on teaching, demonstrations, and machine learning.
A key design innovation is the omission of wires crossing the joints of its limbs, torso, and head. This facilitates continuous rotation and contributes to improved reliability by reducing the potential for wire breakage. Atlas’s hands feature three digits designed for versatility, capable of adjusting to different configurations to function as a three-fingered hand or with one digit repositioned as a thumb. Tactile sensors on the fingers provide crucial data to its neural network for precise object manipulation.
Broader technological shifts have contributed to the recent evolution of humanoid robotics, including deep learning and computer vision utilizing powerful GPUs, advanced actuation through proprioceptive electric motors (transitioning from heavy hydraulics around 2016), and the adaptation of large language models (LLMs) for autonomous planning of complex, multi-step tasks.
Advanced Training Methodologies
Atlas's impressive capabilities are a result of sophisticated training methods:
- Supervised Learning: Machine learning scientist Kevin Bergamin demonstrated this by using a virtual reality headset to directly control the humanoid. This process guides the robot through tasks, generating invaluable data for training its AI models for autonomous execution.
- Motion Capture: Human movements, such as jumping jacks, are meticulously recorded using motion capture body suits. This data is then fed into Boston Dynamics' machine learning process to teach the robot natural movements.
- Simulation Environments: A crucial component of training involves extensive simulations. Over 4,000 digital Atlas avatars underwent six hours of training in simulation environments. These simulations introduced challenges like slippery floors or stiff joints to optimize the robot's performance of specific actions. Optimized skills are subsequently uploaded into the AI system for all Atlas robots.
Current Capabilities and Unmatched Agility
Since 2021, Atlas has demonstrated an significantly expanded range of motion and agility. It can perform a diverse set of actions including running, crawling, skipping, dancing, jumping jacks, and cartwheeling. Its advanced design allows for 360-degree rotation of its arms, head, and torso. The robot is also capable of self-righting from a prone position using only its feet.
Playter stated that Atlas's advanced range of motion aligns with the company's objective to develop robots that can surpass human physical limitations. For example, Atlas can rotate its upper torso 180 degrees to change direction without reorienting its lower body, showcasing an agility beyond human biomechanics.
Remaining Challenges and Limitations
Despite considerable progress since 2015, when bipedal robots faced significant stability and locomotion challenges, experts indicate that fundamental difficulties persist. Current Atlas models cannot reliably handle all sets of stairs or doorways, highlighting the complexities of navigating varied real-world environments.
Kuindersma noted that Atlas is not yet proficient at routine daily tasks such as dressing or pouring coffee. Challenges also remain in refining teleoperation systems, particularly regarding the precise control of gripper shape, motion, and force for dexterous manipulation.
Future Outlook and Industry Implications
The humanoid robot market is poised for significant growth, with Goldman Sachs projecting it to reach an estimated $38 billion within the next decade. U.S. robot manufacturers, including Boston Dynamics, are part of this competitive market, which also includes state-supported Chinese companies. Playter acknowledged the technical lead of U.S. firms but also identified potential competition from the scale of investment by international competitors.
Playter estimates it will be several years before Atlas becomes a full-time worker at Hyundai. He predicted that humanoids would fundamentally alter the nature of work, with robots undertaking repetitive and physically demanding labor. He also indicated that these robots would still require human management, construction, training, and servicing. Benefits cited include enhanced strength, heat tolerance, and the ability to operate in hazardous environments. While advancements in AI and software can occur rapidly, the development and deployment of reliable and affordable robotic hardware are expected to require a considerable timeline.