Robots are system-level systems that consist of sensors that sense their surroundings, actuators that act on their perceptions, and calculations that process data to make coherent and consistent responses to their applications. Robotics is to a large extent the art of systems integration, both from a software and a hardware perspective. Previous studies have revealed that as much as 70% of the resources in robotics are spent on integration rather than developing final applications. Only recently, as lower-end industrial robots move towards mass adoption, have robotics companies started to focus on software development on top of the hardware. Nonetheless, robots are highly specialized systems designed to perform a range of tasks with high reliability and precision. Therefore, the relationship between the hardware and software capabilities within the robot is very important.
Most robots exchange information and meet timing requirements through internal networks. In this sense, robots belong to time-sensitive networks.
Adaptive robots are those that can successfully cope with new situations. To be an adaptive robot worthy of the name, it must have at least one of three basic characteristics: adaptive behavior, adaptive mechatronics, and adaptive computing. Robots with all three basic features can be regarded as "fully adaptive robots".
In summary, robots are highly specialized systems capable of performing a range of tasks with high reliability and precision. The relationship between the hardware and software capabilities inside the robot is very important. Therefore, it is very important to choose a suitable computing platform for the robot system. The platform should be able to simplify system integration and meet power consumption requirements, while at the same time adapting to the changing needs of robotics applications.
AdapTIve Mechatronics is a concept that has been around for decades. Gosselin approaches this topic from a mechanical point of view and defines an adaptive system as one that can successfully respond to new conditions. In addition, he also defined the adaptive robotic mechanical system as an adaptive system, which can adapt to new external conditions by highly relying on mechanical properties. Gosselin offers prototype examples of a variety of purely mechanical adaptive robotic systems, including adaptive robotic hands. In addition, Ivanov also studies adaptive robots from a purely mechanical point of view. In his research, he proposes that adaptive robots should consider the use of adaptive electric drive devices, changing the output of the electric drive devices according to the load. He named this adaptive behavior self-regulation and believes that adaptive electric drives can achieve high energy efficiency in robotic applications. This research is extended to the field of mechatronics through the use of sensor input control mechanisms, resulting in a study of adaptive robot control using sensor feedback. Examples include visual feedback or force sensor feedback.
Adaptive robot behavior is not a new concept. It can be traced back to the behavior-based robot approach proposed by Brooks and his subsumpTIon architecture in the mid-1980s. In the 1990s, different research groups were working on ways to improve robot autonomy by making robots flexible and self-organizing. They often do so through some kind of control mechanism that uses some form of artificial neural network connected to the robot's sensors and actuators. Later Ziemke summarized this approach in his research and proposed the concept of adaptive neural robots.
This concept creatively pioneered the use of the term "adaptive robot," referring to the control of autonomous agents using artificial nervous systems and adaptive technologies. In his recent RoboTIc album Robotic Fabrication in Architecture, Art and Design 2018, the author cites the term "adaptive robots" as Fully adaptive behavior by adding perception and processing (as opposed to Ivanov's concept of "adaptive robots", which focuses on mechanical construction). Adaptive robots are robots that can adapt to changing environmental conditions and material properties while maintaining a degree of predictability through the addition of sensors. Mayoral-Vilches et al., who was contemporaneous with this study and had the same idea, proposed the concept of an autonomous adaptive robot. The concept uses hardware modularity and artificial intelligence (a reference to the 1990s trend of adaptive neural robots) to save the effort and time needed to build such robots.
Adaptive computing is the third basic feature of adaptive robots. For robots, it refers to the ability of the robot to adapt one or more properties of its computing system (such as determinism, power consumption, or throughput) during operation. As in "What is Adaptive Computing? FPGA is an ideal technology to realize adaptive computing. FPGA was first proposed by Freeman, one of the co-founders of Silas in 1984, which laid the foundation for adaptive computing. FPGA has versatility and powerful function, high efficiency, and low cost. Because other processing architectures are implemented on FPGAs and parallel processing is possible, FPGAs can be used to handle almost any task inside a robot. Another feature that makes FPGAs adaptive computing is the ability to tailor data path widths and register lengths to the needs of each robot application.
Examples of how adaptive computing can be used in robots include pipelining for accelerated motion planning, distributed synchronization, or time-sensitive elastic communication.
