Humanoid Robotics Technology
Tencent Robotics X Lab has officially unveiled its latest development, the human habitat robot Xiaowu. Packed with innovative technologies and capabilities, Xiaowu is designed to function as a general-purpose robot that seamlessly integrates into human environments, aiming to live in harmony with people.
As the latest generation of robot developed by Tencent Robotics X Lab, Xiaowu integrates the core capabilities of previous generations of robots, integrating a number of self-developed technologies such as a four-legged wheel-foot composite design, a large area of tactile skin, multi-fingered dexterous hands, and safe human-machine physical interaction. Experiments have shown that Xiaowu can complete actions such as walking and carrying objects in a real human living environment, handle complex tasks, and interact with people.
The main body design is inclusive: the four-legged wheel-foot composite design allows the robot to walk faster and more steadily
The biggest difference between Xiaowu and previous robots is that it uses a composite design of four straight legs and wheels and feet on the main body, which not only maintains the obstacle-crossing ability of a legged robot, but also maintains the operating efficiency of a wheeled robot, allowing the robot to run faster on flat ground.
In terms of motion mode, on flat ground, Xiaowu can switch between four-wheel mode and two-wheel mode. In the four-wheel mode, Xiaowu can expand and reduce the support space according to different load requirements and environmental space through the design of cross legs. On uneven terrain, Xiaowu uses retractable straight legs with active force control to achieve the effect of spring suspension, thereby ensuring the stability of the upper body. In scenes such as stairs and curbs, Xiaowu can use a wheel-foot combination to walk over smoothly .
In terms of load capacity, Xiaowu effectively improves the load-bearing capacity through the straight leg design. Each straight leg can be extended or shortened independently, and the support height can be adjusted according to different working space requirements, thereby realizing a wide working space of “reaching high above and touching the ground below”, such as helping users to pick up and place objects in high places, or operate in low spaces.
In addition to its motion mode and load capacity, Xiaowu can also fold and unfold autonomously. When it is not needed to work, Xiaowu can fold up autonomously to reduce space occupied and facilitate transportation.
In addition to the innovation in form, Xiaowu uses a self-developed dual-encoder high-torque density actuator to meet the demand for large waist loads, making the robot more powerful. At the same time, a self-developed large curved surface covered tactile skin with 180 detection points is deployed on the arms, which enables Xiaowu to have more acute “senses” and respond accordingly to changes in the external environment.
Unified control framework: stronger perception, more flexible control
Faced with diverse and complex tasks in human-living environments, Xiaowu needs to have both agile mobility and dexterous operation capabilities, which places high demands on the control of the robot.
Based on a unified control framework, Xiaowu realizes terrain recognition and precise positioning by sensing environmental information, thereby achieving autonomous adaptation to multiple terrains, full-body motion control that integrates movement and operation, and stable, efficient, and sensitive movement capabilities in all modes.
The unified framework of Xiaowu is roughly divided into three modules. The first is high-precision visual perception positioning and state estimation in large-scale complex environments. Tencent Robotics X Lab has built a high-frequency, low-latency, high-precision real-time online positioning and mapping (SLAM) system based on sensors such as lidar and IMU, and further integrated it with the wheel-foot composite odometer, which allows Xiaowu to better perceive the environment and its own state.
Secondly, with multi-level (path level-trajectory level) online motion planning, Xiaowu can use sensor data such as lidar and RGBD camera to detect dynamic and static obstacles in the surrounding environment in real time, and plan the optimal path and control instructions online to avoid collisions and accidents, ensuring that tasks are completed safely and efficiently in complex environments, such as actively avoiding obstacles when pushing a wheelchair with a person.
Finally, there is the whole-body motion control for integrated mobile operations that adapt to multiple terrains. If encountering terrain with obstacles, such as stairs that are common in human living environments, Xiaowu will move to the vicinity of the stairs, adjust its own mode from the four-wheel motion mode to the four-legged standing mode, and then pass through the stair terrain. In the process of climbing the stairs, Xiaowu can combine the precise step terrain around it and its own state estimation to plan the reachable foothold positions on the steps and generate a series of subsequent center of mass trajectories. Subsequently, the whole-body motion controller will solve the control instructions for each joint based on the robot’s whole-body dynamics model, center of mass trajectory, and foothold information, and control Xiaowu to climb the stairs. After climbing the stairs, Xiaowu can switch from the four-legged stepping mode back to the four-wheel motion mode, making it easier to complete other complex and diverse tasks on flat ground in a more efficient and energy-saving posture.
Furthermore, Xiaowu has the ability to actively control the legs, which is similar to the active suspension function of a car. It can adapt to various complex terrains, such as flat ground, slopes, wavy slopes, cobblestone ground and other common household and municipal terrains, without the need to modify the existing human living environment. In this mode, Xiaowu estimates the ground contact point in real time based on the force sensor and wheel-foot odometer, and uses adaptive impedance control and whole-body motion control algorithms to enable the robot to adapt to different terrains. In addition, the joint control module uses an adaptive model-free friction compensation algorithm to compensate for friction in real time, allowing it to move smoothly at low and high speeds, improving the accuracy and stability of force control.
Safe multimodal human-robot physical interaction: making robots more friendly
Relying on tactile and visual perception, as well as recognition, planning and control algorithms, Xiaowu is able to conduct safe and comfortable physical interactions with humans, and accurately understand, predict and meet people’s daily mobility needs, such as assisting the elderly to walk and sit.
Safe and comfortable human-machine physical interaction depends on the robot’s in-depth understanding and modeling of human needs, status, physical interaction processes, and its own capabilities.
Taking holding and supporting as an example, in order to help the elderly complete this daily action, Xiaowu needs to comprehensively consider the characteristics of the elderly’s movement process from sitting to standing, the elderly’s physical condition (such as motor ability and physical needs), Xiaowu’s own physical ability and other issues .
To achieve this goal, Tencent Robotics X Lab proposed an optimal control model for the robot’s holding and supporting process. The model takes into account factors such as the elderly’s own movement ability (such as joint movement range, driving ability, etc.), body structure (such as height, weight, etc.), movement characteristics of the standing process, and load distribution of the robot’s holding and supporting process.
In addition, in order to make the model more in line with the personalized movement characteristics of the elderly, the team collected a small amount of data on the elderly being held and supported, and allowed the model to learn the optimal model parameters that are suitable for the elderly being held and supported, so that the model is more in line with the movement characteristics of the elderly being held and supported to stand.
In addition, the multimodal perception system is also one of the key modules . It gives Xiaowu human-like perception capabilities, enabling it to more accurately identify humans and complex living environments.
Among these perception modules, the visual and tactile systems play a vital role. The visual system enables Xiaowu to quickly locate and recognize the state of humans, so that it can interact and respond more effectively.
In the process of holding and supporting the elderly, the high-resolution tactile system can help Xiaowu provide support and assistance in a safer and more precise way, effectively avoiding unnecessary pressure or harm to the elderly. The integration of these technologies not only improves Xiaowu’s functionality, but also enhances its sensitivity and humanized performance in actual operation.
SOURCE: Tencent
