A local obstacle avoidance method for mobile robots in partially known environment
Introduction
Local obstacle avoidance is a fundamental problem in autonomous navigation for mobile robot. Most of the navigation approaches in partially known environment combines a global navigation method to find a feasible free path leading to the goal and a local navigation method to follow the path avoiding obstacles. Several well-known local obstacle avoidance methods work by a direction for the robot to head in, in Cartesian space or configuration space [1], [2]. For example, the Artificial Potential Fields methods (APF) [3], [4] were first proposed by Khatib [5]. The main idea is to generate attraction and repulsion forces, within the working environment of the robot, to guide it to the goal. The goal point has an attractive influence on the robot and each obstacle tends to push away the robot. Potential field methods provide an elegant solution to the path-finding problem. Since the path is the result of the interaction of appropriate force fields, the path-finding problem becomes a search for optimum field configurations instead of the direct construction of an optimum path. APF uses vector sums of repulsive and attractive virtual force to compute a desired robot heading. The velocity of robot is proportional to the magnitude of the potential vector. The Vector Field Histogram method (VFH) [6], [7] uses a two-dimensional Cartesian histogram grid as a world model, which is updated continuously with range data sampled by on-board range sensors. The VFH method subsequently employs a two-stage data-reduction process in order to compute the desired control commands for the vehicle. In the first stage the histogram grid is reduced to a one-dimensional polar histogram that is constructed around the robot’s momentary location. Each sector in the polar histogram contains a value represents the polar obstacle density in that direction. In the second stage, the algorithm selects the most suitable sector from among all polar histogram sectors with a low polar obstacle density, and the steering of the robot is aligned with that direction. Robot’s velocity chosen after the direction has been selected is proportional to the distance obstacle ahead. While this method produces smoother travel and can handle both narrow and wide openings, it does not account for the fact that when the robot turns they typically move along arcs, rather than in straight lines. Further more, it is still not adequate to deal with vehicle dynamics, which can cause problems in cluttered environments.
Many approaches have been addressed to deal with local obstacle avoidance problem in dynamic environment. Those works were classified into two categories in terms of the knowledge of the movement of obstacles [8]. In the first category, movements of obstacles are completely unknown to the robot [9]. The collision-free near-optimal paths for mobile robots were generated purely depending on the information acquired by the sensor on board. In the second category, movements of obstacles are completely known. In recent studies some methods are presented for optimal motion planning where a start to goal trajectory is computed at discrete time intervals by searching a tree of feasible avoidance maneuvers [10]. Literature [11] investigated behavior of a single mobile robot which is navigated by an “iterated forecast and planning” scheme in an environment where multiple obstacles are moving around. This navigation scheme searches a feasible path for a robot in () space by a heuristic method. The movement of each obstacle is then forecasted under the assumption that it moves with a piecewise constant velocity.
Velocity space approaches choose angular velocity along with linear velocity, and can take vehicle dynamics into account. The Curvature Velocity Method [12] formulates the local obstacle avoidance problem as one of constrained optimization in the velocity space of the robot. Constraints that result from the robot’s physical limitations and the environment are placed on the velocity space of the robot. The robot chooses velocity commands that satisfy all the constraints and maximize an objective function that trade off speed, safety and goal-directedness. The method has been proved to be efficient and real-time in partially known environments indoors. However, CVM converts Cartesian space to configuration space through expanding the radii of the obstacles by the radius of the robot. Subsequently, it calculates the tangent curvature by the geometric center of the obstacle. Obviously, it is not easier to achieve space conversion than calculate the tangent curvature by the obstacle’s center without prior knowledge of obstacles in partially known or unknown environments. Though CVM produces reliable, smooth and speedy navigation in office environments, it often fails to guide the robot into an open corridor towards the goal direction and it sometimes lets a robot head towards an obstacle until the robot gets near the obstacle. These problems derived from the fact that CVM chooses commands based on the collision-free length of the arcs assumed to be robot’s trajectories. It is easy to ignore the case that the most appropriate path perhaps exists in short distance.
Ko and Simmons [13] realized these limitations and developed the lane curvature method This method combines CVM with a new directional method called the lane method, which divides the environment into lanes, and then chooses the best lane to follow to optimize travel along a desired heading. A local heading is then calculated for entering and following the best lane, and CVM uses this heading to determine the optimal linear and angular velocities. The lane method chooses a direction to a wide and collision-free opening since it decides heading direction based on the collision-free distance and the width of the lane. While VFH method chooses a direction to the opening with wide collision-free angular range rather than an opening with big width. So VFH often forces a robot into a narrow opening near the robot because even a narrow opening can offer wide collision-free angular range if the opening is close to the robot. In this respect, the lane method can provide safer heading commands to CVM than VFH. Nevertheless, because the lane calculation modifies the sensor vision of the environment, the robot may not see clear space with enough distance free of collision caused by the lack of radial projected vision. In practical experiments, the sharp turns that caused by delays in finding clear spaces have been observed.
Fernandez and Sanz [14] combine a new directional method, called beam method (BM), to improve the performance of CVM as well as LCM. The beam method calculates the best heading that will be delivered to CVM to obtain the optimal linear and angular velocities. The resulting combined technique is called the beam curvature method (BCM). BM obtains a divergent radial projection model of the environment based on the sensors’ common position. The projection model is defined by a set of sectors or radial beams. This model is simplified and then a set of possible candidate beams is determined. After that, the best beam is calculated by maximizing an objective function. The local heading target is calculated by imposing security constraints around the best beam. Experiments in different scenarios indicate that BCM can not only find the opening faster, but also produce smoother trajectory than CVM and LCM. However, BCM assumes that there are no openings behind the obstacles, which may leads to the false selection of optimal opening when appropriate path lies in between the two obstacles especially in clustered environment. Furthermore, BCM performs well in clustered environment with static obstacles, but may fail to avoid a moving obstacle in dynamic environments. It is because BCM itself doesn’t take the corresponding velocity of the obstacles in robot’s local reference frame into account. Therefore, we used a prediction model to help BCM realize local obstacle avoidance in dynamic environment.
The remainder of this paper is organized as follows. A method that converts Cartesian space to configuration space is introduced in Section 2. The improved BCM has been given in Section 3. Section 4 describes the prediction model for BCM in detail. Some experimental results of autonomous navigation are summarized in Section 5. Finally, conclusions are given in Section 6.
Section snippets
Conversion from Cartesian space to configuration space
Although various methods about obstacle avoidance have been thoroughly considered in the literature, it is evident that not all the developed methods are able to give an appropriate answer to all possible situations. In most partially known environment, the global pose of the robot in world reference frame and the detailed shape of obstacles are unknown, the robot has to detect and localize the obstacles by sensors on board. But the navigation system can supply the robot with the angle error
Formation, classification and selection of beams
BCM uses a two-step approach to realize navigation in partially known or unknown environment. First, it searches for the best beam by maximizing an object function. Second, the best beam is delivered to CVM to produce linear and angular velocity commands that range from physical constraints of the robot.
The prediction model of collision
The velocity space approach, including CVM, LCM and BCM, treat the navigational problem as one of constrained optimization in velocity space. Varied experiments prove that these methods may produce reliable, smooth and speedy navigation in office environments. However, there exists a common shortcoming in all these methods that the robot often fails to avoid a moving obstacle in cluttered environments. Furthermore, though LCM often produces unnecessary delays in finding clear spaces and thus
Experiments and results
In this section, we examine the proposed method by a series of navigation experiments in partially known environments with static or moving obstacles. The task of the robot is to search for a known object with its camera and stop in front of it. In the experiment, Pioneer III, equipped with laser range finder and visual system, is used as a hunter. The other three robots, named HIT II, act as the moving obstacles. Both the robot and the obstacle’s velocity is not more than 50 cm/s.
In order to
Conclusions
According to the experimental results in different scenes, we can draw conclusions as follows:
First, we present a new method to convert Cartesian space to configuration space so that the calculation of tangent curvature and the formation of beams need not rely on the detail shape of obstacles around the robot. This improvement makes all the velocity space approaches available in autonomous navigation in unknown or partially known environment.
Second, different from the original beam method, the
Acknowledgements
This work had been completely supported by the Foundation Project for Scientific Development from Nanjing University of Science and Technology (grant number: XKF09015) and partially supported by Special funds for the Research on Scientific Creation and Talent People (grant number: 2008RFQXG067). We would like to thank all the researchers who contribute themselves to the velocity space method for their great influences on this work.
Chaoxia Shi, born in 1974, received BA and MA degrees from Laiyang Agricultural College and Northeast Dianli University in 1997 and 2003 respectively. Since 2003, he has been a Ph.D. degree candidate in Harbin Institute of Technology and received the DA degree in 2007. He is now a instructor in Nanjing University of Science and Technology and his research interests include autonomous navigation for mobile robot and pattern recognition.
References (18)
- et al.
Development of a new minimum avoidance system for a behavior-based mobile robot
Fuzzy Sets and Systems
(2009) - et al.
Motion planning for mobile robots in a time-varying environment
Journal of Robotics and Mechatronics
(1996) - et al.
Dynamic motion planning for mobile robots using potential field method
Autonomous Robots
(2002) - et al.
A voronoi diagram-visibility graph-potential field compound algorithm for robot path planning
Journal of Robotic Systems
(2004) - O. Khatib, Real-time obstacle avoidance for manipulators and mobile robots, in: Proc. IEEE Intl. Conference on Robotics...
- et al.
The vector field histogram—fast obstacle avoidance for mobile robots
IEEE Journal of Robotics and Automation
(1991) - et al.
Evolutionary programming-based univector field navigation method for fast mobile robots
IEEE Transaction on Systems, Man, and Cybernetics—Part B: Cybernetics
(2001) - et al.
Robot motion planning in dynamic environments with moving obstacles and target
International Journal of Mechanical Systems Science and Engineering
(2007) - S. Ishikawa, A method of indoor mobile robot navigation by using fuzzy control, in: Proc. IEEE/RSJ Int. Workshop on...
Cited by (46)
Obstacle avoidance in dynamic environments based on velocity space optimization
2020, Robotics and Autonomous SystemsCitation Excerpt :Artificial Potential Field (APF) methods [24] create a potential landscape and use gradient descent for local navigation and have been extended to avoid dynamic obstacles [25–27]. Shi et al. [28] developed a local obstacle avoidance method that combines the prediction model for collisions with an improved BCM that can avoid moving obstacles in dynamic environments. The prediction model can forecast the obstacle’s future trajectory when the robot moves along a new curvature decided on by using a certain beam and thus avoid colliding with the obstacle, assuming that the obstacle keeps its current velocity.
Admissible gap navigation: A new collision avoidance approach
2018, Robotics and Autonomous SystemsCitation Excerpt :Other research efforts were directed towards incorporating the motion constraints into the obstacle avoidance problem, selecting an admissible velocity rather than a steering direction. The outcome of these efforts was the development of various methods [30–32], known later as velocity space methods. However, it was the Dynamic Window Approach DWA [33,34], initially proposed in [35], that won popularity among the scientific community.
A collision-free motion planning method by integrating complexity-reduction SLAM and learning-based artificial force design
2018, Robotics and Autonomous SystemsTangential Gap Flow (TGF) navigation: A new reactive obstacle avoidance approach for highly cluttered environments
2016, Robotics and Autonomous SystemsCitation Excerpt :Other common techniques take the dynamic constraints of the robot into account and choose a steering command rather than a moving direction. The Curvature Velocity method (CVM) [24,25], and the Dynamic Window (DWA) approach [26,27] are the most popular ones. They work by adding constraints, coming from physical limitations and sensory data, to the velocity space, and choose the speed that satisfies all constraints and maximizes an objective function.
Chaoxia Shi, born in 1974, received BA and MA degrees from Laiyang Agricultural College and Northeast Dianli University in 1997 and 2003 respectively. Since 2003, he has been a Ph.D. degree candidate in Harbin Institute of Technology and received the DA degree in 2007. He is now a instructor in Nanjing University of Science and Technology and his research interests include autonomous navigation for mobile robot and pattern recognition.
Yanqing Wang, born in 1977, is an instructor in Harbin University of Science and Technology. Her main research interests are image processing and pattern recognition.
Jingyu Yang, born in 1941, received the BS degree in computer science from Nanjing University of Science and Technology (NJUST), Nanjing, China. He is currently a professor and chairman in the Department of Computer Science at NJUST. He is the author of more than 300 scientific papers in computer vision, pattern recognition, and artificial intelligence. His current research interests are in the areas of pattern recognition, robot vision, image processing, data fusion, and artificial intelligence.