This paper describes a humanoid robotics platform (DRC-HUBO+) developed for the Defense Advanced Research Projects Agency Robotics Challenge (DRC) Finals. This paper also describes the design criteria, hardware, software framework, and experimental testing of the DRC-HUBO+ platform. The purpose of DRC-HUBO+ is to perform tasks by teleoperation in hazardous environments that are unsafe for humans, such as disaster zones. We identified specific design concepts for DRC-HUBO+ to achieve this goal. For a robot to be capable of performing human tasks, a human-like shape and size, autonomy, mobility, manipulability, and power are required, among other features. Furthermore, modularized joints and a user-friendly software framework were emphasized as design concepts to facilitate research on the robot tasks. The DRC-HUBO+ platform is based on DRC-HUBO-1 and HUBO-2. The torque of each joint is increased compared to that in DRC-HUBO-1 owing to its high reduction ratio and air-cooling system. DRC-HUBO+ is designed with an exoskeletal structure to provide it with sufficient stiffness relative to its mass. All wires are enclosed within the robot body using a hollow shaft and covers to protect the wires from external shock. Regarding the vision system, active cognition of the environment can be realized using a light-detection and ranging sensor and vision cameras on the head. To achieve stable mobility, the robot can transition from the bipedal walking mode to the wheel mode using wheels located on both knees. DRC-HUBO+ has 32 degrees of freedom (DOFs), including seven DOFs for each arm and six DOFs for each leg, and a solid and light body with a height of 170 cm and a mass of 80 kg. A software framework referred to as PODO, with a Linux kernel and the Xenomai patch, is used in DRC-HUBO+.

Research on a terrain-blind walking control that can walk stably on unknown and uneven terrain is an important research field for humanoid robots to achieve human-level walking abilities, and it is still a field that needs much improvement. This paper describes the design, implementation, and experimental results of a robust balance-control framework for the stable walking of a humanoid robot on unknown and uneven terrain. For robust balance-control against disturbances caused by uneven terrain, we propose a framework that combines a capture-point controller that modifies the control reference, and a balance controller that follows its control references in a cascading structure. The capture-point controller adjusts a zero-moment point reference to stabilize the perturbed capture-point from the disturbance, and the adjusted zero-moment point reference is utilized as a control reference for the balance controller, comprised of zero-moment point, leg length, and foot orientation controllers. By adjusting the zero-moment point reference according to the disturbance, our zero-moment point controller guarantees robust zero-moment point control performance in uneven terrain, unlike previous zero-moment point controllers. In addition, for fast posture stabilization in uneven terrain, we applied a proportional-derivative admittance controller to the leg length and foot orientation controllers to rapidly adapt these parts of the robot to uneven terrain without vibration. Furthermore, to activate position or force control depending on the gait phase of a robot, we applied gain scheduling to the leg length and foot orientation controllers, which simplifies their implementation. The effectiveness of the proposed control framework was verified by stable walking performance on various uneven terrains, such as slopes, stone fields, and lawns.

We developed a software system for operating the humanoid robot DRC-HUBO+ in disaster circumstances that the US Defense Advanced Research Projects Agency suggested. This system was developed under the consideration of the stability of whole system, the systemic software environment for multiple developers, and the recovery routine when the system encounters a seriously abnormal situation. With these design goals, we devised our strategy for three domains: system, vision, and communication. Following the strategy we made two software frameworks (PODO and VPC) and one logical flow of data for remote process management. With this software system, we conducted all Tasks in the DARPA Robotics Challenge Finals, and we won the competition.

In this letter, we propose learning-based force control of twisted string actuators (TSAs) using a neural network-based inverse model. A learning-based force controller is designed using the input and output data of TSAs without a dynamic model of TSAs. Furthermore, the neural network-based inverse model is utilized to reduce model errors and handle nonlinearities between the inputs and outputs of the TSAs. The trained neural network-based inverse model is directly implemented as a force controller for the TSAs. Additionally, we propose data collection methods utilizing three types of inputs to improve the performance of the proposed controller. To verify the improved performance resulting from the proposed data collection methods, we compared the performance of the learning-based force controller for each dataset in the TSA hardware. We then selected the dataset with the best performance among the proposed inputs through experiments. Additionally, to verify the performance of the learning-based force controller, a reference force tracking experiment was performed and compared with a proportional-integral-derivative (PID) controller and feedback linearization. The learning-based force controller utilizing the selected input-based dataset demonstrated higher force tracking performance than the other controllers. Consequently, the TSA's learning-based force control demonstrates robust force control performance.