Published on: February 25, 2025 / update from: February 25, 2025 - Author: Konrad Wolfenstein
From stumbling to somersault in robotics: AI-upgrade defines humanoid skills-Image: Xpert.digital
The future of the humanoids: Atlas is through Reinforcement Learning Smarter
Strategic partnership: Boston Dynamics optimizes Atlas for real applications
In an announcement, Boston Dynamics, a pioneer in the field of dynamic robot, and the Robotics & AI Institute (RAI Institute), a research institution under the direction of the renowned robotic expert and former CEOs of Boston Dynamics, Marc Raibert, announced a strategic partnership. The declared goal of this cooperation, which officially found its start in February 2025, is the significant improvement of the skills of the advanced humanoid robot atlas by using reinforcement learning (reinforcing learning). This cooperation promises not only to make Atlas more flexible and agile, but also to qualify it for a wider spectrum of real applications and thus pave the way for a new era of humanoid robotics.
Suitable for:
Core targets of future -oriented cooperation
The partnership between Boston Dynamics and the RAI Institute focuses on a number of ambitious goals that aim to transform the fundamental skills of Atlas and to develop it from an impressive research demonstrator into a varied and practical tool. At the center of these efforts are three main areas:
The bridging of the Sim-to-Real gap: the way from the simulation to reality
One of the greatest challenges in robotics, especially in the area of Reinforcement Learning, is the transfer of skills learned in simulations to the real world. Simulations offer an ideal environment for training robots because they enable unlimited amounts of data, complete control over the environment and the possibility of simulating dangerous or cost -intensive scenarios with risk -free. Robots can carry out countless iterations of movements and tasks in virtual worlds without the risk of damage or injuries.
The reality, on the other hand, is much more complex and unpredictable. Physical robots operate in a world full of sensory noise, unforeseen disorders, inaccuracies in modeling and the constant challenge of variability. What works in a perfectly controlled simulation can fail in chaotic reality. The "Sim-to-Real-Lücke" describes exactly this discrepancy.
The partnership between Boston Dynamics and the RAI Institute has set itself the goal of closing this gap using innovative methods and algorithms. The researchers are working on developing robust and generalizable movements that work reliably not only in simulation, but also in the real world. This includes the development of advanced simulation environments that map the physical reality more precisely, as well as the use of techniques such as domain randomization and adaptive simulation in order to make the models trained in simulations more resistant to the imponderables of the real world. The success in this area is crucial to exploit the full potential of Reinforcement Learning for robotics and use robots in real, unstructured environments.
Improvement of loco manipulation: the art of movement and interaction
The ability to locate loco manipulation, i.e. the simultaneous transportation and manipulation of objects, is a key ability for robots that should act in complex and dynamic environments. Imagine a humanoid robot that moves through a warehouse to pick packages, or a robot that eliminates debris in a disaster zone and at the same time searches for survivors. In all of these scenarios, it is essential that the robot not only moves efficiently, but can also interact with its surroundings at the same time.
However, the development of advanced Loko manipulation strategies is an enormous challenge. It requires close coordination between movement planning, rail planning, gripping planning and the level of strength. The robot must be able to adapt its movements and manipulations to the constantly changing conditions of its surroundings in real time.
As part of the partnership, the researchers will develop new and innovative strategies to raise ATLAS loco manipulation skills to a new level. This includes researching algorithms for the simultaneous planning and gripping planning, the development of robust power control strategies for manipulation of various objects and the integration of sensory information into the control loop to enable reaction fast and adaptive loco manipulation. The improvement of loco manipulation is a crucial step to make Atlas a really versatile and useful tool for a variety of applications.
Research into full-body contact strategies: the synergy of poor and legs
Humanoid robots like Atlas have the unique potential to move and interact in a way that is very similar to human movement. This ability to integrate the entire body, including the arms, legs and fuselage, into complex movements and tasks, opens up completely new opportunities for robotics. All-body contact strategies go beyond simple manipulation with the arms and use the synergy between the arms and legs to enable high-performance movements and tasks.
Think of a person who carries a heavy object. He not only uses his arms, but also his legs, his fuselage and his entire body to stabilize the weight, to keep the balance and to transport the object efficiently. Similarly, humanoid robots should be able to use their entire body to manage complex tasks that require close coordination between the arms and legs.
The researchers focus on the development of advanced regulatory algorithms and planning strategies for high -performance full body movements and tasks. This includes areas such as dynamic running, jumping, climbing, lifting and carrying heavy objects, manipulation in cramped rooms and the interaction with complex environments. The research of full-body contact strategies is of crucial importance in order to exploit the full potential of the humanoid form factor and to develop robots that can move and interact in the world in a natural and intuitive way.
The importance of this directional cooperation
The partnership between Boston Dynamics and the RAI Institute is of immense importance for robotics and AI research community for several reasons. First, she combines two leading organizations in the field of robotics, each with unique strengths and skills. Boston Dynamics is known worldwide for its impressive and dynamic robot platforms such as Atlas, Spot, Handle and Stretch. The RAI Institute under the direction of Marc Raibert brings decades of experience in the development of top technologies for intelligent machines and in the use of Reinforcement Learning to complex robotics problems.
Marc Raiber, the founder of the RAI Institute, is an icon of robotics. As a former Boston Dynamics CEO, he has significantly shaped the development of the company and produced some of the most impressive robots in the world. Robotics research has had a lasting impact on his vision of robots, which can move as clever and versatile in the real world as people and animals. With the founding of the RAI Institute, Raiber continues his mission to expand the limits of the possible in robotics and AI.
The collaboration is based on a solid basis of earlier joint projects, including the "Reinforcement Learning Researcher Kit" for the four -legged robot spot. This KIT enables researchers worldwide to develop and test Reinforcement Learning algorithms on the Spot platform. The successful development and implementation of this kit has shown that both organizations are able to work together effectively and to develop innovative solutions in the field of re -forcement learning for robotics.
By using Reinforcement Learning to Atlas, one of the most advanced and powerful humanoid robots in the world, partners expect significant progress in the development of humanoid skills. Reinforcement Learning offers the potential to train robots, to manage complex tasks that would be difficult to implement with traditional programming approaches. It enables robots to learn through interaction with their surroundings, to adapt and to continuously improve their skills.
Boston Dynamics and the RAI Institute have undertaken to publish regular updates and demonstrations of their work with Atlas in order to make progress in humanoid robotics accessible to the general public. This transparency is important to strengthen trust in robotics and AI research and to promote social acceptance for these technologies. The planned publications will not only inform the scientific community, but also inspire the public for the fascinating possibilities and challenges of humanoid robotics.
Joint research and development in detail
The cooperation between Boston Dynamics and the RAI Institute is divided into several core areas of research and development, which are closely linked and complement each other:
Development of a common re-follow-up learning training pipeline for Atlas
At the center of the partnership is the development of a state-of-the-art Reinforcement learning training pipeline, which is specially tailored to the needs and skills of Atlas. This pipeline will form the basis for training dynamic and generalizable behavior for mobile manipulation. It includes all steps of the Reinforcement learning process, from the definition of reward functions and the selection of suitable algorithms to the development of simulation environments and data acquisition to validation and transfer of the learned behavior on the real robot.
The training pipeline will be modular to ensure flexibility and adaptability to various tasks and environments. It will integrate advanced techniques of Reinforcement Learning, such as Deep Reinforcement Learning, Model-Based Reinforcement Learning and multi-agent Reinforcement Learning to maximize the efficiency and robustness of the training. A special focus will be on the development of reward functions that enable Atlas to learn complex tasks without explicitly specifying every step. The reward functions are intended to guide the robot to develop efficient, natural and human -like movements and interactions.
SIM-to-Real transfer: The bridge between virtual and real world
As already mentioned, the SIM-to-Real transfer is one of the biggest challenges in Reinforcement Learning for robotics. The teams will work intensively to bridge the gap between simulations and the real world and ensure that the behavior trained in simulations can be successfully and reliably transferred to the physical hardware.
This requires a multi -layer approach, which includes the improvement of the simulation environments and the development of robust transfer methods. The simulation environments are continuously improved in order to map the physical reality more precisely, including the modeling of friction, contact, inertia and other physical effects. At the same time, techniques such as domain randomization, system identification and adaptive control are used to make the models trained in simulations more resistant to the imponderables of the real world. The goal is to create a seamless transition from the simulation to reality, so that Atlas can use the skills learned in the virtual world without significant loss of performance in real environments.
Focus on key skills for the future of humanoid robotics
The partnership focuses on the development and improvement of key skills that are essential for the practical use of humanoid robots in real environments:
Improved loco manipulation: Handle objects during movement
Atlas is to be able to manipulate objects and devices such as doors, switches, levers, tools and other objects while moving at the same time. This ability is crucial for a variety of applications, from industrial automation to logistics to search and rescue operations. Imagine Atlas, which moves through an rough terrain and at the same time eliminated debris or serves tools to repair a damaged structure.
The improved loco manipulation requires the development of algorithms, which coordinate the movement planning, gripping planning and the level of strength in real time. Atlas must be able to adapt its movements and manipulations to the shape, size, weight and nature of the objects that he manipulates. In addition, he must be able to deal with uncertainties in perception and the surrounding area and to dynamically adapt its plans and movements. The development of these skills will make Atlas a much more versatile and more useful tool for a wide range of applications.
Full body contact strategies: complex movements and heavy loads
The researchers focus on the development of demanding full -body movements that go beyond simple walking and reaching. This includes dynamic running, jumping, climbing, lifting and carrying heavy objects and manipulation in cramped rooms. These skills require close coordination between the arms, legs and fuselage and use the synergy of the entire body to manage complex tasks.
Dynamic running and jumping enable Atlas to move quickly and efficiently in uneven terrain and over obstacles. Climbing extends its range and enables access to difficult areas. Lifting and wearing heavy objects makes him a valuable helper in logistics and construction. Manipulation in cramped rooms enables the use in environments that are difficult to access or dangerous for humans. The development of full-body contact strategies is a crucial step to exploit the full potential of the humanoid form factor and make Atlas a really agile and powerful robot.
Practical implementation and continuous progress control
The partnership between Boston Dynamics and the RAI Institute attaches great importance to a transparent and practice-oriented implementation of your research and development work:
Regular progress reports and demonstrations
Boston Dynamics and the RAI Institute have undertaken to publish periodically progress reports that document the latest developments and success of the cooperation. These reports will not only include written descriptions of progress, but also vivid demonstrations with Atlas, which show the newly acquired skills in action. These demonstrations are published in the form of videos and presentations and made accessible to the scientific community and the general public.
The regular updates and demonstrations serve several purposes. They enable the scientific community to pursue the progress in humanoid robotics and inspire each other. They promote transparency and trust in robotics research and help to increase social acceptance for these technologies. In addition, they offer Boston Dynamics and the RAI Institute the opportunity to receive feedback from the community and adapt their research direction accordingly.
Location of cooperation: Massachusetts, USA
The entire research and development work as part of the partnership takes place in Massachusetts, where both organizations have their headquarters. This spatial closeness promotes close cooperation and direct exchange between the research teams. The teams from Boston Dynamics and the RAI institute work in common laboratories and use the resources and infrastructures of both organizations. This close integration of teams and resources is a crucial factor for the success of the partnership and enables synergies to be used and efficiently promote research and development work.
Expected new skills of Atlas: A look into the future of humanoid robotics
Due to the partnership between Boston Dynamics and the RAI Institute, the Atlas robot is intended to obtain a number of groundbreaking new skills that will make it an even more versatile and useful tool:
Improved mobility and manipulation: agility and precision in motion
Dynamic locomotion
Atlas is to be able to move even more stable and liquid on uneven terrain, in complex environments and even in dynamic scenarios. This includes running, jumping, climbing and the ability to adapt to different surfaces and conditions in real time. The dynamic locomotion is made possible by advanced regulatory algorithms and sensor data fusion that allow Atlas to keep its balance, overcome obstacles and adapt its movements to the respective situation.
Full body manipulation
The robot will implement advanced strategies for full -body contact in order to be able to use, carry, move and manipulate heavy objects precisely and efficiently. This requires a highly developed coordination of arms, legs and fuselage to stabilize the weight, to keep the balance and to handle the objects safely. The full body manipulation will enable Atlas to take on tasks that were previously only reserved for people, such as moving heavy loads in warehouses, on construction sites or in disaster zones.
Advanced environmental interaction: intelligent interaction with the world
Object manipulation
Atlas should learn to manipulate a variety of objects and devices in its area, including doors, switches, levers, valves, tools, containers and much more. This ability will enable him to act in human environments and perform tasks that require interaction with the existing infrastructure. Object manipulation requires advanced perception skills to recognize, locate and identify objects, as well as sophisticated gripping and manipulation strategies in order to handle them safely and efficiently.
Adaptability to materials and structures
The robot will be able to automatically and intelligently adapt its strength, speed and movements to different materials and structures without damaging or destroying them. This is crucial for the safe and reliable interaction with the real world, in which robots will encounter a variety of surfaces, materials and objects. The adaptability is achieved through the use of strength and torque sensors, tactile sensors and advanced regulatory algorithms that enable Atlas to monitor and adapt its interactions in real time.
Learning ability and generalization: The basis for future innovations
More efficient learning through Reinforcement Learning:
The use of advanced purforcement learning techniques is intended to enable Atlas to learn new skills much faster and more efficiently than before. This includes the development of algorithms that accelerate learning, the data
Suitable for:
Your global marketing and business development partner
☑️ Our business language is English or German
☑️ NEW: Correspondence in your national language!
Konrad Wolfenstein
I would be happy to serve you and my team as a personal advisor.
You can contact me by filling out the contact form or simply call me on +49 89 89 674 804 (Munich) . My email address is: wolfenstein ∂ xpert.digital
I'm looking forward to our joint project.