Sheng Bi

Affordance Research in Developmental Robotics: A Survey

Affordances capture the relationships between a robot and the environment in terms of the actions that the robot is able to perform. The notable characteristic of affordance-based perception is that an object is perceived by what it affords (e.g., graspable and rollable), instead of identities (e.g., name, color, and shape). Affordances play an important role in basic robot capabilities such as recognition, planning, and prediction. The key challenges in affordance research are: (1) how to automatically discover the distinctive features that specify an affordance in an online and incremental manner and (2) how to generalize these features to novel environments. This survey provides an entry point for interested researchers, including: (1) a general overview; (2) classification and critical analysis of existing work; (3) discussion of how affordances are useful in developmental robotics; (4) some open questions about how to use the affordance concept; and (5) a few promising research directions.

Multi-objective optimization for a humanoid robot walking on slopes

A multi-objective gait optimization method for a humanoid robots walking on slopes is proposed in this paper. Firstly, based on SCUT-I humanoid robot and slopes model, the complicated process of walking on slopes is parameterized so that hip and ankle trajectories are planned based on IPM (inverted pendulum model). Secondly, a multi-object optimization method based on walking stability, velocity and energy is proposed to search for the optimal gait parameters that are step, walking cycle, the supporting time of double feet and the height of foot. Finally, SCUT-I robot model is built and simulated by Matlab6.5, which verifies the effectiveness and stability of this multi-objective gait optimization method that makes the robot walk on slopes more stably, quickly and lower energy consumption.

Outdoors real-time information analysis system for elderly based on smart phone

Aging population is growing, and the young people are more concerned about the real-time condition of the elderly. This paper aims to propose a real-time monitoring system, obtaining and analyzing the outdoors dynamic information of the elderly, which is composed of three cooperative modules (mobile client, server and sensor). The sensor is equipped with a Bluetooth module to the data transmission, and the server is mainly responsible for distinguishing between the falling attitude and normal activity and record the outdoors motion path of the elderly. The result is displayed on the client, including the mobile phone and browser. If the abnormal condition emerges, the phone of the elderly immediately alarms and the server sends message to ask their children and relevant departments for help at once. The system shows a good real-time performance and an accurate result.

Real-time gesture recognition system based on Camshift algorithm and Haar-like feature

This paper propose a real-time combined method of Camshift [1] algorithm and Haar-like feature detection [2] for tracking and recognizing hand gesture in images acquired by a possibly moving camera. A Haar-like classifier is used during the initializing of the system to acquire the users hand color. Camshift algorithm is applied with the acquired color to track the position of the hand, accompanied with a two-dimensional Kalman filter to prevent target occlusion. An offline training with Adaboost learning algorithm [3] of various hand gestures is employed to allow the classifier to recognize static hand gestures. With the information provided by Haar-like classifier, Tracking with Camshift will have more robust performance and less likely to lost targets. As the experimental result shows, the method we proposed have better performance than simply applying any of the two methods, especially in some complicate background conditions where skin-color disturbances or occlusion exist.

A Gesture Recognition Method Based on Binocular Vision System

This paper demonstrates a gesture recognition approach based on binocular camera. The binocular vision system can deal with stereo imaging problem using disparity map. After the cameras are calibrated, the approach uses skin color model and depth information to separate the hand from the environment in the image. And the features of the gestures are extracted by feature extraction algorithm. These gestures as well as their features constitute a set of training examples in machine learning. The Support Vector Machine (SVM), which is supervised learning models, are used to classify these gestures that are labeled with their meaning, such as digits gesture. In training and classification processes, we use the same feature extraction algorithm handling the gesture image and SVM can recognize the meaning of a gesture. The gesture recognition method mentioned in this paper represents a high accuracy in recognizing number gestures.

A biped humanoid robot's gait planning method based on Artificial Immune Network

A biped humanoid robot model with 12 degree of freedom is developed in this paper. To facilitate the gait pattern planning, the 3D inverted pendulum model and the ZMP are introduced to enable a human-like stable walking. Since the searching of best walk primitive is a multi-objective optimization problem, a modified aiNet Algorithm as well as SGA Algorithm is applied to the optimization process. Finally, the control parameters worked out by both algorithms are verified and compared in simulation. We find out that the result of aiNet provides the robot with better stability than SGA while they are similar in mobility.

Gesture recognition based on Kinect

With the rapid development of computer science, gesture recognition has been a highlight of research in the area of Human Computer Interaction (HCI). Generally speaking, gesture recognition can be divided into two types: static gesture recognition and dynamic gesture recognition. Under the background of the aging population, how to use the method of gesture recognition to help the elderly adapt to “Intelligent Age” is a meaningful issue, which deserves more attention. This paper describes a gesture recognition method based on Kinect, a 3D somatosensory camera sensor. This method involves skeleton tracking, where the skeleton data is produced from the depth images obtained via Kinect. Extensive experiments demonstrate the superior performance of the proposed methods over Kinect.

Multi-feature gesture recognition based on Kinect

Human Computer Interaction (HCI) has been a popular research area during the last few years. Compared with the tradition HCI methods such as using a keyboard or mouse, people prefer to have their tasks done in a more natural way. As an essential form of non-verbal communication in daily life, gesture is a good choice to turn the ideas into reality. Although various recognition methods are proposed to solve the problem, these methods are time-tensed, space-tensed or miscellaneous. This paper introduced a new method to recognize the hand gesture correctly and efficiently. The recognition is done through two phases: the skeleton phase concerning capturing and processing skeleton feature of the hand gesture, and the hand phase focusing on extracting hand contour feature of the hand gesture. Experimental results confirm an overall 94% accuracy in recognizing and matching the pre-defined templates and robustness to backgrounds.

Automatic feature extraction and optimal path planning for robotic drawing

Automatic robotic drawing is a fantastic demo to show the combination of intelligence and robot technique. It requires automatic feature extraction, complex robot path planning and optimization, which can make many contributions in industrial manufacturing. In this paper, we propose a hybrid method by combining local binarization with global binarization to extract features. The extracted features which will be drawn by a robot refer to some cells with complex shapes. The path of drawing should be optimized because the drawing process is time consuming. Therefore, these cells need to be partitioned into multiple cells due to its complexity. The path of drawing each cell can be generated separately, and the trajectories of drawing all cells need to be connected to form a complete trajectory. Here we propose a new solution to optimize the connection while considering the acceleration and deceleration of the robot tool. All these algorithms have been implemented and attractive result is achieved by experiment. The proposed method can be used in many industrial applications such as painting and grinding.

Biped walking on rough terfrain using reinforcement learning

In this paper, we propose a novel reinforcement learning method to stabilize biped walking on rough terrain. For the state space and the action space of the biped walking problem is continuous, the neural network is used in our method, which is based on actor-critic learning, to approximate the policy function of actor and the value function of critic. The neural network learns on-line through the process. The proposed method is examined in simulation. The simulation results show that the robot can learn to improve the stability of walking on rough terrain by using the proposed method.