Bio-Inspired Robots: Merging AI with Real-World Applications

[Un article de The Conversation écrit par Olivia Chevalier – Ingénieur de recherche, Institut Mines-Télécom Business School – Gérard Dubey – Sociologue, Institut Mines-Télécom Business School – & Johann Héraut – Maître-assistant en robotique bio-inspirée, Laboratoire des Sciences du Numérique de Nantes, IMT Atlantique – Institut Mines-Télécom]

In robotics, the use of such approaches suggests rapid progress in the empowerment of humanoid robots. The boom in computer vision, based on these new architectures of neural networks, has, for example, made it possible to considerably improve the interaction of robots with their environment, in particular to avoid obstacles and to manipulate objects. Nevertheless, a limit remains to the advances of the AI ​​in robotics: humanoid robots still struggle to reach the fluidity and precision of human movements, in particular with regard to beer -on and grip.

Indeed, the coordination of the motor functions necessary for movement is not just a mechanical planning, comparable to a succession of blows in a game of failures. In reality, the human movement and, more broadly, the animal movement is based on a complex tangle of operations and interactions involving internal components to the individual, such as motor control (the equivalent of AI in the robot), the sensory system or biomechanics, as well as external components, such as physical interactions with the environment.

For example, an amateur jogger is capable of maintaining its overall stable gaze despite the irregularities of the terrain and fatigue, by taking advantage of passive properties of the human body (from the plantar articulation to the movement of the hips), reflexes, as well as an engine control of the eye and cervical muscles. Our musculoskeletal and nervous systems have thus evolved jointly to meet the challenges posed by heterogeneous and unpredictable environments.

In comparison, to accomplish tasks that require continuous adjustment between action and its objective, robots have a limited number of actuators (in other words, engines) and even more of sensors.

In this context of material constraints, can we really hope that the computation power of AI and their learning capacities are enough to achieve the motor performance observed in humans and in animals?

The so -called “embodied” approach precisely takes the opposite of the purely calculatory approach by not dissociating the algorithmic and physical components of the robot. On the contrary, it aims to explore possible synergies between the body and control, between passive and active mechanisms, so that a “motor intelligence” or “embodied” also emerges from these interactions. This article thus examines the limits and perspectives of synergies between artificial intelligence, the robot and its environment.

Towards autonomous robots: two phases in the history of robotics

Rodney Brooks, former Director of the AI ​​laboratory at Massachusetts Institute of Technology (MIT), has led a research program entitled: “The Cog Project: Building A Humanoid Robot” for years. Brooks distinguishes two phases in the history of robotics research. During the first phase (years 1970-1980), research is based on the fact that the robot program contains the data from the environment in which it evolves, or rather where it does not evolve. During the second phase, from the 1990s, research was based precisely on interaction with the environment.

This dynamic relationship to the environment makes it possible to see to what extent the robots are complicated and self-organized, or are self-employed over the history of robotics research. As Brooks says, “humanoid intelligence requires humanoid interactions with the world”. It is therefore a question of developing programs capable of changing themselves according to interactions with the environment.

The second robotics, or how AI systems can make robots more autonomous

The research of the second robotics are therefore aimed at developing a ” Behaviour-Based Robot “(Robot based on a behavioral model), one of the requirements of which interests our words: so that the action of the robot is close to ours, we must, among other things, suppose it” not planned “.

It is, precisely, first of all that progress in AI is fruitful. But to what extent can AI make it possible to reduce the gap between the behavior of robots and those, extremely complex, which we seek to make them reproduce? Because AI plays a big role in the design of robots, in the manufacture of materials they are made of and obviously in simulation and modeling, it offers the means of this embodied approach.

One of the main objectives of this approach is the autonomy of robots, that is to say their ability to make decisions and adapt to their environment.

To better understand this point, we can oppose the physicist approach to that of embodied AI. Thus, the traditional approach (also qualified as “physicist” or “objectivist”) does not give the means to know if a machine can feel Or to understandwhile the embodied AI approach poses the problem of machine autonomy in terms which would in principle allow to verify this hypothesis of the possibility for a machine to feel or understand. Indeed, considering, on the one hand, that the whole – the body – is more than the addition of the parts (the components) and, on the other hand, than the phenomena that interest us (phenomenal consciousness, understanding, sensation, for example) are the emerging product of this All Immersed in the environment, this second approach offers the means to test this hypothesis.

Flexible robotics (in its bio-inspired version) seems to be more fit than other robotic approaches mentioned above to get closer to this objective of the embodied approach. Indeed, inspired by the behaviors of biological organisms and trying to reproduce certain aspects, it aims to build robots which adapt to the environment and build their autonomy in their interaction with it.

Another imagination of the relationship between humans and machines

The coupling of robotics and AI potentially foreshadows another imagination of the relationship between humans and machines and the technique to nature than that which prevailed in the industrial era.

Indeed, from the 1940s, cybernetic theory, with the concept of “homeostasis” (self -regulation of the organism with its environment), at the sources of the current AI, was already a thought of the insertion of machines into the environment. The cybernetic association between sensors and signal processing had paved the way for bringing together mechanical intelligence (which can be briefly defined as intelligence mainly governed by algorithms) with that of living beings in the natural world. The autonomy of the machines was however always thought of the model of the capacity of living organisms to maintain their internal balances by resisting the disturbances of the environment, that is to say by granting priority to everything that makes it possible to reduce this external “disorder”.

Current research in robotics seem to influence this report by considering that the disturbances of the environment represent potentialities and clean resources which deserve to be understood and apprehended as such.

It is not only a question today of inserting a robot into a neutral environment or already known by him, but of making this environment – unpredictable, often unknown – a component of his behavior. This research is thus focused on the interactions of the body or the mechatronic system with the physical world-that is to say with the contact forces and the process of information processing implemented in the sensitive experience by living beings.

Soft robotics, soft robotics, bio-inspired, embodied intelligence are possible variations of these new approaches and reveal the importance of the role played by AI in the opening of robotics to other issues than those which were traditionally its own, by bringing lighting or by raising certain scientific locks.

The new robotics therefore does not only lead to a renewal of interest in the living. The conceptions of the machine which it carries-a machine immersed in its environment, which depends deeply-resonate strongly with the new approaches during the living in biology which define it mainly from its interactions. The new dialogue that is established between robotics and biology thus contributes to rethinking the boundaries that separate the living from non-living.

Therefore, could the embodied approach to robotics make it possible to fill the gap between machine and living?