The quadruped robot FROG-I (Four-legged Robot for Optimal Gaits), designed and built by our group, is about 1150mm long, 700mm wide, and 950mm tall, with a total weight of approximately 55kg. Each leg contains one hip pitch joint and one knee pitch joint, both of which are powered by DC motors. Still, FROG-I possesses one passive compliant prismatic DOF at each toe. An embedded controller performs control, sensing and and communicates with a host computer through a wireless connection. The main specifications of FROG-I are summarized as follows.
For the traditional CPG controller, parameters couple with each other strongly, and it is not easy to tune the ideal signals used for gait control. Based on the general formulation of the solution to CPG, we developed a controller for quadruped locomotion , whose parameters have explicit physical meaning compared to the traditional CPG controller. The proposed controller inherits the advantages of the traditional CPG controller, and successfully solves the problems of fixed phases, non-zero start points and hind leg dragging, which exist in the traditional CPG controller. We implemented the proposed controller on both a simulated quadruped model and the real quadruped platform FROG-I.
It is well recognized that mammals can utilize different gait patterns to locomote over a variety of support surfaces. Quadruped robot can also use collections of gait patterns, such as walk, trot, pace, bound, gallop and so on. This research is focused on better understanding the gait transition of quadruped robots and designing the transition algorithms that allow the robot switch smoothly from one gait to another. In our algorithms, we change the CPG parameters for a specific time to generate different control curves for the new locomotion gaits.
Quadruped robots are typically high degree-of -freedom systems, inherently under-actuated, and have changing contacts and constraints with respect to the support surface. These features complicate the inverse dynamics computation and control in real time, which are crucial techniques for quadruped locomotion in natural environments.
Future Work will address the following issues for quadruped locomotion:
• Presenting the general representation for inverse kinematics and inverse dynamics
• Reducing the computation of the inverse kinematics and inverse dynamics
• Presenting the inverse dynamics control strategy
• Implementing compliant walking control on the real quadruped platform
• Evaluating the inverse dynamics control
Trajectory Planning
This research is motivated by the desire to allow the quadruped locomotion on inaccessible or discontinuous terrain. Our current efforts are targeting developing trajectory planning and control techniques to allow capabilities of jumping obstacles and crossing ditches. A recently developed trajectory planning algorithm was examined that enables FROG-I to overstride a wood board (videos:overstriding1.wmv, overstriding2.wmv).
Future Work will address a set of problems:
• Real time perceiving for discontinuous terrain
• Global/Local visual servo
• Optimal path planning for robot legs and feet
• Trajectory planning for legs
Geology Adaptation Locomotion
It has been learned from experimental studies on the biological locomotion that some mammals show marvelous adaptation capability of locomotion on different environment substrates by regulating ground (substrate) reaction forces (GRF). Although kinematics and dynamics of the quadruped robot are distinct from those of quadruped animals in their mechanisms, actuators, sensors and so on, we try to enhance the ability of the quadruped robot to traverse in a variable geology environment.
Future work will be aimed at testing our proposed method in natural environments, such as ice, sand, snow, etc.
Autonomous Locomotion with Stereo Vision
Autonomous locomotion in unstructured environments is a great challenge for quadruped robots, especially when there is little knowledge about the environments. Currently, for some quadruped robots, the walking tasks are established and the knowledge of terrains is known previously. We propose an effective motion control system with stereo vision for autonomous locomotion over unknown complicated terrain, including the following themes:
• Disparity estimation method based on SAD and post-processing algorithm in severe conditions
• Terrain modeling in global coordinate systems according to quadruped robots' kinematics and attitude
• Visual angle calculation with RANSAC algorithm
• Grid map generation
• Cost map construction
• Real-time motion planning algorithm based on multiple gaits and gait transition
Future work will be focused on dealing with autonomous walking in dynamic environments, such as ground planes with moving obstacles.
Embedded Controller
In order to improve the real-time control ability, we have developed an embedded controller for FROG-I. The controller includes two parts, the high-level controller and the low-level controller. The high-level controller based on the Intel Xcale PXA270 communicates with the PC operation platform via wireless connection. The low-level controller based on TMS320DM642 and TMS320f2812 controls the actuated joints, and processes signals from cameras, angle sensors, foot contact sensors and gyroscope. In addition, the high-level controller receives the information from the video processor and guides the low-level controller to perform proper actions. With this design of control architecture, the controller can process the signals and control the locomotion effectively.
Contact
Dr. Wei Wang Institute of Automation
Chinese Academy of Sciences
95 Zhongguancun Donglu Road
Beijing, 100190, China
The quadruped robot FROG-I (Four-legged Robot for Optimal Gaits), designed and built by our group, is about 1150mm long, 700mm wide, and 950mm tall, with a total weight of approximately 55kg. Each leg contains one hip pitch joint and one knee pitch joint, both of which are powered by DC motors. Still, FROG-I possesses one passive compliant prismatic DOF at each toe. An embedded controller performs control, sensing and and communicates with a host computer through a wireless connection. The main specifications of FROG-I are summarized as follows.
For the traditional CPG controller, parameters couple with each other strongly, and it is not easy to tune the ideal signals used for gait control. Based on the general formulation of the solution to CPG, we developed a controller for quadruped locomotion , whose parameters have explicit physical meaning compared to the traditional CPG controller. The proposed controller inherits the advantages of the traditional CPG controller, and successfully solves the problems of fixed phases, non-zero start points and hind leg dragging, which exist in the traditional CPG controller. We implemented the proposed controller on both a simulated quadruped model and the real quadruped platform FROG-I. 
It is well recognized that mammals can utilize different gait patterns to locomote over a variety of support surfaces. Quadruped robot can also use collections of gait patterns, such as walk, trot, pace, bound, gallop and so on. This research is focused on better understanding the gait transition of quadruped robots and designing the transition algorithms that allow the robot switch smoothly from one gait to another. In our algorithms, we change the CPG parameters for a specific time to generate different control curves for the new locomotion gaits. 
Quadruped robots are typically high degree-of -freedom systems, inherently under-actuated, and have changing contacts and constraints with respect to the support surface. These features complicate the inverse dynamics computation and control in real time, which are crucial techniques for quadruped locomotion in natural environments. 
This research is motivated by the desire to allow the quadruped locomotion on inaccessible or discontinuous terrain. Our current efforts are targeting developing trajectory planning and control techniques to allow capabilities of jumping obstacles and crossing ditches. A recently developed trajectory planning algorithm was examined that enables FROG-I to overstride a wood board (videos:
It has been learned from experimental studies on the biological locomotion that some mammals show marvelous adaptation capability of locomotion on different environment substrates by regulating ground (substrate) reaction forces (GRF). Although kinematics and dynamics of the quadruped robot are distinct from those of quadruped animals in their mechanisms, actuators, sensors and so on, we try to enhance the ability of the quadruped robot to traverse in a variable geology environment.
Autonomous locomotion in unstructured environments is a great challenge for quadruped robots, especially when there is little knowledge about the environments. Currently, for some quadruped robots, the walking tasks are established and the knowledge of terrains is known previously. We propose an effective motion control system with stereo vision for autonomous locomotion over unknown complicated terrain, including the following themes:
In order to improve the real-time control ability, we have developed an embedded controller for FROG-I. The controller includes two parts, the high-level controller and the low-level controller. The high-level controller based on the Intel Xcale PXA270 communicates with the PC operation platform via wireless connection. The low-level controller based on TMS320DM642 and TMS320f2812 controls the actuated joints, and processes signals from cameras, angle sensors, foot contact sensors and gyroscope. In addition, the high-level controller receives the information from the video processor and guides the low-level controller to perform proper actions. With this design of control architecture, the controller can process the signals and control the locomotion effectively.
