International Journal of Humanoid Robotics

We developed a bipedal robot equipped with brake and clutch mechanisms to change the number of active and passive joints, thereby enabling various types of movements including normal active walking using 12-dof joints, under-actuated walking using brake, and passive-based walking using clutch and passive joints. In this paper, we describe three technologies to achieve the proposed system and show experimental results on active and semi-passive walking. The first technology comprises a small and high-strength clutch mechanism to sustain the massive weight of life-sized robots using actuators for joint and dog clutch control. The second technology comprises a walking controller using a simulation-based optimization technique to consider passive joint dynamics instead of depending on the inverse kinematics problem, thereby enabling the control of the under-actuated leg. The last technology is model parameter identification to achieve unstable passive-based walking in real-world considering the body as well as environmental parameters such as ground slope. To the best of our knowledge, the proposed robot is the first to achieve both active and passive-based walking using a bipedal body. This enables the implementation of the passive-walking technology to active-joint robots and expands the application possibility of passive joint for bipedal robots.

Unsteady locomotion and the dynamic environment are two problems that block humanoid robots to apply visual Simultaneous Localization and Mapping (SLAM) approaches. Humans are often considered as moving obstacles and targets in humanoid robots working space. Thus, in this paper, we propose a robust dense RGB-D SLAM approach for the humanoid robots working in the dynamic human environments. To deal with the dynamic human objects, a deep learning-based human detector is combined in the proposed method. After the removal of the dynamic object, we fast reconstruct the static environments through a dense RGB-D point clouds fusion framework. In addition to the humanoid robot falling problem, which usually results in visual sensing discontinuities, we propose a novel point clouds registration-based method to relocate the robot pose. Therefore, our robot can continue the self localization and mapping after the falling. Experimental results on both the public benchmarks and the real humanoid robot SLAM experiments indicated that the proposed approach outperformed state-of-the-art SLAM solutions in dynamic human environments.

Humanoid robots are often supposed to share their workspace with humans and thus have to deal with objects used by humans in their everyday life. In this article, we present our novel approach to humanoid navigation through cluttered environments, which exploits knowledge about different obstacle classes to decide how to deal with obstacles and select appropriate robot actions. To classify objects from RGB images and decide whether an obstacle can be overcome by the robot with a corresponding action, e.g., by pushing or carrying it aside or stepping over or onto it, we train and exploit a convolutional neural network (CNN). Based on associated action costs, we compute a cost grid containing newly observed objects in addition to static obstacles on which a 2D path can be efficiently planned. This path encodes the necessary actions that need to be carried out by the robot to reach the goal. We implemented our framework in the Robot Operating System (ROS) and tested it in various scenarios with a Nao robot as well as in simulation with the REEM-C robot. As the experiments demonstrate, using our CNN, the robot can robustly classify the observed obstacles into the different classes and decide on suitable actions to find efficient solution paths. Our system finds paths also through regions where traditional motion planning methods are not able to calculate a solution or require substantially more time.

This paper presents a balance-centered planner for object re-posing. It uses Center-of-Mass (CoM) constraints to preserve robot stability and provides stable, IK-feasible, and collision-free upper-body poses, allowing the robot to complete dexterous object manipulation tasks with different objects. The technical contributions of the planner are two-fold. First, it evaluates the robot stability margin for each robot pose during manipulation planning to find a stable manipulation motion. Second, it provides an RRT-inspired task-related stability estimation used to compare different bipedal stances. Simulations and real-world experiments are performed with the HRP-5P humanoid robot, the 5th generation of the HRP robot family, to validate the planner and compare different robot stances and approaches for object re-posing. The experiment results suggest that the proposed planner provides robust behavior for the humanoid robot while performing re-posing tasks.

The human hand is a complex, highly-articulated system, which has been the source of inspiration in designing humanoid robotic and prosthetic hands. Understanding the functionality of the human hand is crucial for the design, efficient control and transfer of human versatility and dexterity to such anthropomorphic robotic hands. Although research in this area has made significant advances, the synthesis of grasp configurations, based on observed human grasping data, is still an unsolved and challenging task. In this work we derive a novel, constrained autoencoder model, that encodes human grasping data in a compact representation. This representation encodes both the grasp type in a three-dimensional latent space and the object size as an explicit parameter constraint allowing the direct synthesis of object-specific grasps. We train the model on 2250 grasps generated by 15 subjects using 35 diverse objects from the KIT and YCB object sets. In the evaluation we show that the synthesized grasp configurations are human-like and have a high probability of success under pose uncertainty.

A three-finger under actuated robotic hand with dexterous force control and inherent compliance is developed and tested. A simplified biomimetic finger design is generated and applied with mechanical intelligence principles carefully designed and embedded such that optimal trajectories for grabbing are naturally followed and the fingers can automatically conform to the goal object. A generalizable potential energy flow theory is then proposed to explain the mechanism behind the mechanical intelligence. The theory is also supported by experimental results. Quasi-direct drive actuators were developed to actuate the robotic hand with proprioceptive force sensing and inherent compliance. The hand performs delicate force-controlled manipulation with a simple compliance controller implemented.