International Journal of Humanoid Robotics

This paper presents a new control approach for a humanoid biped robot to perform highly dynamic locomotion and keep balance under large external disturbances. The proposed control approach mainly contains two parts: a simplified model-based task planner and an online whole-body controller. The periodic stability is achieved with a foot placement policy presented in this paper. The landing position is predicted and adjusted at every time step to adapt to uncertain disturbances. Then the desired task trajectories are tracked with a quadratic programming (QP) based task space whole-body controller, while satisfying some specific physical constraints. The whole-body controller outputs the optimized joint accelerations, which are then used to produce desired joint torques for the robot’s actuators. The positions of the center of mass (CoM) along sagittal and lateral axes are not controlled in the whole-body controller. Meanwhile, the torques at the ankles are constrained to be small and the robot follows a passive-like behavior, which is energetically efficient and helps the feet of the robot to fit better in the environment. Simulations on several highly dynamic gaits have been conducted, two examples among which are presented in this paper. The first demonstrates a walking gait with an increasing sagittal speed and its capability to resist large external disturbances. In the second example, the robot is able to run stably on uneven terrain without knowing any information about it.

This paper investigates the influence of reference motion quality and other design choices on the performance of deep reinforcement learning for bipedal walking with Proximate Policy Optimization (PPO). We use parametrized Cartesian quintic splines to generate reference actions for an omnidirectional walk policy. By using parameter sets with different qualities, we show that the performance of the trained policy correlates to the quality of the reference motion. We also show that a policy in Cartesian space performs superior to a joint-space-based one if an advantageous representation of orientation is chosen. Additionally, we show that using an initial bias for the policy speeds up the training and leads to higher performances for policies using position control. We also show that we can achieve a stable omnidirectional walk on a wide variety of simulated humanoid robots.

Self-Balanced lower limb exoskeleton (SBLLE) is designed for rehabilitation and walking assistance for dyskinesia persons with spinal or lower limb muscle injury to regain the locomotion ability in daily activities. In this paper, an innovative feedback control strategy for a fully actuated SBLLE is presented, which helps users to walk stably without crutches or other assistive devices. Exoskeleton is approximated as a simplified center of mass (CoM) model. Based on this simplified dynamic model, the trajectory generation and walking feedback control strategy, which enhances the locomotion stability, is introduced. The proposed strategy is implemented on our exoskeleton, AutoLEE-II, and its stability and dynamic performance are demonstrated in the stable walking simulations and experiments for exoskeleton with a female subject.

Knee-stretched walking is desirable for biped robot because it is a characteristic of human walking as well as lower energy consumption. In this paper, we proposed a new knee-stretched bipedal walking pattern generation approach named Cosine-law-based Spatially Quantized Gait (C-SQG). C-SQG includes two parts, the first one is the Cosine law-based kinematic joint angles generation for knee-stretched biped walking with the hip horizontal displacement; the second one is combining with the concept of Spatially Quantized Dynamics (SQD) and the optimization-based methods. Based on these two approaches, the balanced human-like walking patterns with fully extended knee joints for humanoid robots can be generated. The proposed approach was tested on the CoppeliaSim simulation platform, and the experimental results indicate that the knee-stretched walking patterns for different walking speeds and foot locations can be generated. Compared to the existing knee-stretched gait generation approach, the C-SQG is more robust and adaptive to different desired foot locations appropriately.

A new method is proposed to control bipedal robots to achieve flexible omni-directional motion and robust locomotion under complex disturbances, called the heuristics-based reinforcement learning (HBRL) framework. HBRL shows great training efficiency in simulation. Heuristic reference trajectories play a crucial role in HBRL, which guide the training process. Exploration rewards, leg-foot reset condition, and command curriculum are three significant components to optimize the training process. An estimator network is utilized to supply linear velocities and foot contact information. We train controllers on flat ground in simulation. To demonstrate robustness and versatility, the trained controllers were tested on BRAVER, a point-foot bipedal robot with three joints on each leg. The controllers enabled BRAVER to perform omni-directional locomotion with the maximum forward speed reaching 2m/s. The robot could also maintain balance under external pushing and over uneven terrains.

The robots with humanoid hands can take the place of humans in carrying objects and using tools efficiently in different situations. This paper presents a novel hydraulic-driven robotic hand named the WLRH-II (the second-generation wheel-legged robot humanoid hand) which bears heavy load and has strong robustness. This paper focuses on the design of humanoid hand driving and transmission mechanisms, kinematic and static analyses. First, the linkage mechanism is proposed according to the particle swarm optimization algorithm. Then, the kinematics and the statics of WLRH-II are analyzed in grasp state. In addition, the hydraulic servo control method is used to control the position of a hand through angle feedback. Finally, position tracking and load experiments are completed to verify the reasonableness and effectiveness of the control. The robotic hand can grasp objects of 30kg with 2.85-kg self-weight. The WLRH-II has a high load/self-weight ratio of 10.55 in this paper, which is rare in similar humanoid robotic hands.

Humanoid robots still have a low-level perception of their surroundings, which is a formidable obstacle to performing complex tasks. To improve their perceptual capabilities, high-level semantic information can be incorporated into their perceptual framework so that they can gain the ability to infer and understand their environment accurately. Among the various techniques available, semantic Simultaneous Localization and Mapping (SLAM) becomes a promising avenue for achieving this perception enhancement. While semantic SLAM can enhance a robot’s ability to perform tasks, the accurate perception, and association of semantic objects in complex and dynamic environments remain challenging. Hence, the need arises for a solution that swiftly and precisely associates object measurements with landmarks, considering their motion properties, and promptly rectifies erroneous associations in real time. To this end, we propose a semantic perception approach designed explicitly for dynamic environments, adept at distinguishing between dynamic and static objects. Furthermore, we propose two association strategies: dynamic object association based on semantic map points and static object association based on object pose information. And as the number of object measurements associated with the landmarks increases, we perform the association validation algorithm to verify the association for the landmark to improve the accuracy of the association. The proposed method is extensively evaluated on both simulated indoor sequences obtained from humanoid robot viewpoints and the KITTI dataset. Experimental results show that our approach significantly improves the robustness and accuracy of object association and trajectory estimation.

It is a challenging task to localize and track an in-hand object in robotic domain. Researchers were mainly using the vision as major modality for extracting object’s pose. The vision approaches are fragile when the object is occluded by the robotic arm and hand. To this end, we propose a tactile-based DTI-Tracker (tracking object’s pose via Dynamic Tactile Interaction) approach and formalize the object’s tracking as a filter problem. An Extended Kalman Filter (EKF) is used to estimate the in-hand object pose exploiting the high spatial resolution tactile feedback. Given the initial estimation error, the proposed approach rapidly converges the estimation result to the real pose and the statistic evaluation shows the robustness of the proposed approach. We evaluate this method in physics simulation and real multi-fingered grasping setup while the object is static and movable. The proposed method is a potential tool to foster future research on dexterous manipulation using multifingered robotic hand.