RESEARCH

Real-Time Constrained Nonlinear Model Predictive Control on SO(3) for Dynamic Legged Locomotion

In this work, a novel NMPC framework for dynamic legged locomotion and an efficient algorithm to solve this constrained nonlinear optimization problem are presented. The orientation of the robot among the components that make up the objective function of the optimization problem adopts the manifold configuration of the rotation group. Through the reparameterization, dynamic motions that include configurations in which local parameterizations such as Euler angles suffer singular problems are properly implemented. The computational efficiency obtained through the proposed control framework has given a significant advantage in real-time implementation of dynamic locomotion control of various types of legged robots. The effects are verified by showing robust stabilization in simulation environments, including push recovery and acrobatic situations such as wall-climbing.

Hardware Design Optimization

Cycloid gear based humanoid platform development 

Actuating systems for the proprioceptive control of legged robots must have a high mechanical efciency and mechanical bandwidth for high torque transmission to avoid power transmission losses. We developed a high-efciency compact cycloidal reducer for legged robots that uses needle roller bearings in all parts where contact occurs during the power transfer process inside the reducer, which greatly improves the efciency compared to a cycloidal reducer using free rollers. We also proposed a subcarrier structure that distributes the load and allows the cycloidal reducer to respond robustly to impacts that may occur during the locomotion of the legged robot. The subcarrier increases the stifness, which improves the mechanical bandwidth. A cycloidal reducer was manufactured with > 90% efciency in most operating ranges; it weighs 766 g and can withstand a torque of more than 155 Nm. The cycloidal reducer module coupled with the motor indicated a torque control bandwidth of 72 Hz, so it can be used for legged robots with agile motions.

Hardware Design Optimization

Hydraulic humanoid platform development 

In order to assist with daily tasks such as fetching a beverage, a service robot must be able to perceive its environment and generate corresponding motion trajectories. This becomes a challenging and computationally complex problem when the environment is unknown and thus the path planner must sample numerous trajectories that often are sub-optimal, extending the execution time. To address this issue, we propose a unique strategy of integrating a 3D object detection pipeline with a kinematically optimal manipulation planner to significantly increase speed performance at runtime. In addition, we develop a new robotic butler system for a wheeled humanoid that is capable of fetching requested objects at 24% of the speed a human needs to fulfill the same task.

Robot Intelligence

Automated warehouse management with humanoid robot 

In order to assist with daily tasks such as fetching a beverage, a service robot must be able to perceive its environment and generate corresponding motion trajectories. This becomes a challenging and computationally complex problem when the environment is unknown and thus the path planner must sample numerous trajectories that often are sub-optimal, extending the execution time. To address this issue, we propose a unique strategy of integrating a 3D object detection pipeline with a kinematically optimal manipulation planner to significantly increase speed performance at runtime. In addition, we develop a new robotic butler system for a wheeled humanoid that is capable of fetching requested objects at 24% of the speed a human needs to fulfill the same task.

Dynamic motion design for legged robots traversing complex unstructured environment

  Inspired by the versatile mobility of animals to traverse difficult terrain, researchers have developed many legged robots in the hopes of using them in environments too difficult and dangerous for humans, such as disaster situations, planetary exploration, and battlefields. Situations such as destroyed structures, piles of large obstacles, and rocky cliffs present challenging problems for robots that rely on static locomotion regimes or a limited variety of pre-designed motions, particularly since these situations prevent humans from controlling robot movements on the scene. Current state-of-the trajectory planning algorithms for complex dynamic systems in complex environments easily fail in such scenarios mainly because they (1) require a good initial guess of the robot’s feasible path (often heuristically designed by human operators), (2) involve long calculation times owing to the necessity of making computationally expensive calculations in a high-dimensional search space.

High Speed Running Control of Quadrupedal Robots

  The main objective of the study is to present a controller design scheme which provides a robust running gait of a quadruped robot with the ability to change the running speed over a wide range while handling variations in ground height and stiffness. Drawing on trends in change of stance and swing duration, and vertical impulse of quadrupedal animals as speed increases, a novel algorithm is proposed to prescribe vertical and horizontal impulses to achieve the desired stance and swing duration for different speeds as well as to regulate body pitch dynamics. Application of the control design scheme not only enables MIT Cheetah 2 to run with a wide range of speeds from 0 to 4.5 m/sec (10.1 miles per hour) without change of any control parameters or a further optimization process but also provides 40 cm high dynamic jumping followed by safe landing while running with a speed of 2.5 m/sec (5.5 miles/hour). The proposed impulse planning approach is  extended to the design of high jumping, long distance leaping and rapid turning motions where the jumping height, leaping distance, and turning radius can be precisely controlled as planned.

Learning dynamics of Quadrupedal Robots

Dynamic and agile maneuvers of animals cannot be imitated by existing methods that are crafted by humans. A compelling alternative is reinforcement learning, which requires minimal craftsmanship and promotes the natural evolution of a control policy. However, so far, reinforcement learning research for legged robots is mainly limited to simulation, and only few and comparably simple examples have been deployed on real systems. The primary reason is that training with real robots, particularly with dynamically balancing systems, is complicated and expensive. In the present work, we introduce a method for training a neural network policy in simulation and transferring it to a state-of-the-art legged system, thereby leveraging fast, automated, and cost-effective data generation schemes. 

Humanoid Robot Research Center (Hubo Lab)

Department of Mechanical Engineering
Korea Advanced Institute of Science and Technology
Building N9 #4150

291 Deahak St. Yuseong-gu, Deajeon, Korea

(+82) 042-350-7139