International Journal of Humanoid Robotics

A biologically inspired foveated attention system in an object detection scenario is proposed. Bottom-up attention uses wide-angle stereo camera data to select a sequence of fixation points. Successive snapshots of high foveal resolution using a telephoto camera enable highly accurate object recognition based on SIFT algorithm. Top-down information is incrementally estimated and integrated using a Kalman-filter, enabling parameter adaptation to changing environments due to robot locomotion. In the experimental evaluation, all the target objects were detected in different backgrounds. Significant improvements in flexibility and efficiency are achieved.

In this paper, in order to improve the classification accuracy with features as few as possible, a new hierarchical recognition method based on an improved SVM decision tree and the layered feature selection method combining neural network with genetic algorithm are proposed. The improved SVM decision tree is constructed according to confusion degrees between two emotions or those between two emotion groups. The classifier in each node of the improved decision tree is a SVM. On the emotional speech corpus recorded by our workgroup including 7 emotions, with the features and parameters gotten by the method combining neural network with genetic algorithm, improved SVM decision tree, multi-SVM, SVM-based binary decision tree, the traditional SVM-based decision directed acyclic graph and HMM are evaluated respectively. The experiments reveal that, compared with the other four methods, the proposed method in this paper appears better classification accuracy with fewer features and less time.

Human running can be stabilized in a wide range of speeds by automatically adjusting muscular properties of leg and torso. It is known that fast locomotion dynamics can be approximated by a spring loaded inverted pendulum (SLIP) system, in which leg is replaced by a single spring connecting body mass to ground. Taking advantage of the inherent stability of SLIP model, a hybrid control strategy is developed that guarantees a stable biped locomotion in sagittal plane. In the presented approach, nonlinear control methods are applied to synchronize the biped dynamics and the spring-mass dynamics. As the biped center of mass follows the mass of the mass-spring model, the whole biped performs a stable locomotion corresponding to SLIP model. Simulations are done to obtain a repeatable hopping for a three-link underactuated biped model. Results show that periodic hopping gaits can be stabilized, and the presented control strategy provides feasible gait trajectories for stance and swing phases.

This paper aims at the guarantee of yaw moment requirement for the ZMP convention. In order to compensate for the yaw moment of whole body during motion, we present two motion generation methods for the upper body of the humanoid robot. The ground reactive torque and the effect of arms swinging on whole body locomotion are analyzed. Therefore trunk spinning and arms swinging methods are planned to improve the robot motion stability, based on compensating for the yaw moment. These two methods are further compared with each other from the viewpoint of energy consumption. The performance and the feasibility of the proposed methods are evaluated by the simulation results.

Passive dynamic walking has been developed as a possible explanation for the efficiency of the human gait. In this paper, we investigate the effects of foot shape on energetic efficiency and dynamic stability of passivity-based bipeds with upper body. Three walking models with point feet, round feet and flat feet were presented. Each model has an upper body constrained to keep midway between legs. We use computer simulations to find which foot shapes are indeed optimal in view of energetic efficiency and dynamic stability for a passive dynamic biped with upper body. Simulation results indicate that feet improve both the energetic efficiency and dynamic stability of passive dynamic bipeds.

This paper describes and compares three different approaches to estimate simultaneous localization and mapping (SLAM) in dynamic outdoor environments. SLAM has been intensively researched in recent years in the field of robotics and intelligent vehicles, many approaches have been proposed including occupancy grid mapping method (Bayesian, Dempster-Shafer and Fuzzy Logic), Localization estimation method (edge or point features based direct scan matching techniques, probabilistic likelihood, EKF, particle filter). In this paper, a number of promising approaches and recent developments in this literature have been reviewed firstly in this paper. However, SLAM estimation in dynamic outdoor environments has been a difficult task since numerous moving objects exist which may cause bias in feature selection problem. In this paper, we proposed a possibilistic SLAM with RANSAC approach and implemented with three different matching algorithms. Real outdoor experimental result shows the effectiveness and efficiency of our approach.