Robotic involvement is envisaged for exploration of human-inaccessible areas such as planetary space, confined and unstructured environments, and radioactive places. An exploration mission usually includes multiple tasks that are difficult or even impossible to finish using a single robot. Modular robots aim to solve this problem by providing a robotic system wherein robotic modules can be reconfigured to accomplish diverse tasks.
In this work, research is undertaken on the design, manufacturing and control of a modular robotic system consisting of straight extending modules. Each robotic module of the modular robot can be actively controlled or can respond passively to external forces. The modular elements can be connected simply for ease of manual reconfiguration.
A new connectivity strategy for building modular robotic structures using rigid connector nodes, active and passive modular elements is investigated. Comparisons of the new connectivity and a conventional connectivity using compliant connector nodes are made with respect to kinematics, locomotion and deformation of some robotic structures. Modular units including a prismatic actuator, a rigid connector node and a passive revolute joint are then designed, manufactured and tested. More modular elements are further replicated for building modular robotic structures leading to a final prototype system with eight prismatic actuators, four rigid connector nodes and four passive revolute joints. Each prismatic actuator is equipped with a locking mechanism and possesses three different working states: it can either be actuated, locked or passive. The three-state prismatic actuator is self-contained with its own computation, communication, actuation and sensing capabilities.
A proportional-integral-derivative (PID) controller is implemented to control the position of the prismatic actuator. The actuation and locking forces of the prismatic actuator are experimentally evaluated. The prismatic actuator can vertically lift an external load of 29.4 N. The locking force of the mechanical locker is 78.6 N, enabling the actuator to be capable of vertically supporting a weight of about 2.5 kg in the locked state. The minimum force required to passively move the prismatic actuator is also measured as 8.34 N. The performance of the PID controller, three states and state transitions of the prismatic actuator are then validated by a series of physical experiments. Experimental results demonstrate that the maximum absolute value of the displacement error is to be 0.175 mm in the actuated state, and state transitions between actuated, locked and passive states are physically achievable. Moreover, state transitions of two and multiple prismatic actuators are also realized resorting to communications between the prismatic actuators.
As a high-level control strategy, a central pattern generator (CPG) neural network is first applied to modular robotic structures composed of the fabricated robotic modules. Physical experiments show that the modular robotic structures achieve a worm-like locomotion gait through the coordination of their actuators' movements, substantiating the feasibility and effectiveness of the mechanical design and control strategy. Modular robotic structures with greater number of elements are constructed in a physics-based robot simulator. A generalized CPG neural network and a role-based control method are developed for controlling these simulated modular robots. Computer simulations are then conducted to further demonstrate locomotion capability of modular robotic structures composed of three-state prismatic actuators. Simulation results show that the generalized CPG method is scalable to a broad range of robotic structures with different number of modules. The three-state prismatic actuator can be applied to releasing physical constraints of a robotic structure during task execution and achieving a walking pattern by using state transitions.
