Another field that cannot be satisfied with a procedural approach is robotics. Robots live in a changing environment and must constantly adapt to unforeseen events. It is through sensors that robots perceive this environment. The BT will therefore articulate the robot's abilities to move, to interact to achieve the objective assigned to it.
We are talking about a physical robot. But the concept can be extended to bots, to autonomous agents collaborating with each other in a cyber space. For example, ROS, for Robot Operating System, which is defined as the OS of robots, includes BTs at the heart of its system. Those who want to go further can consult the ROS site, or the university productions of Colledanchise & Ogren.
Use cases of behavior trees
BTs are therefore an implementation of the adaptive approach . Their strength lies in adaptation and modularity . This branch of AI to which BTs belong is today in the shadow of deep learning, which is based on the massive use of data and has enjoyed resounding success for about ten years.
However, in the context of digital transformation, it would be interesting to rethink certain business processes to benefit from the power offered by an AI driven by adaptive logic . Indeed, deep learning sometimes suffers from its magical effect; it is difficult to explain the results obtained. This drawback, which is sometimes prohibitive, disappears in the context of BT where the rules are ordered and facilitate the explanation of the results obtained.
Use case: Chatbot
At Klee Group, we use BTs to make our chatbot " intelligent ". A chatbot is an agent that must answer questions. To do this, the bot starts a conversation and will ask additional questions managed by the BT in order to provide a reliable answer to the requester. The flexibility of the BT makes it possible to guide the conversation based on a knowledge base and the user's context.
A "classic" chatbot is described in a procedural form with multiple sequential decision trees. Opting for an adaptive approach with behavior trees offers two major benefits:
Efficiency: End users will not have any unnecessary questions. If you have ever called a telephone support, you have certainly been treated to an operator, who coldly "goes down" a decision tree without taking into consideration your context. This leads to questions unrelated to your problem. The behavior tree works differently. It is driven by the objective and thus avoids superfluous questions. Its objective is to quickly answer the question asked (think of GPS!).
Adaptation: A chatbot is a living digital being, in the sense that it must continually consider new questions. The very nature of behavior trees makes it easy to modify and add to adapt to users. The work for the chatbot administrator will, de facto, be simplified.
Use case: Configuration
Another use case concerns the choice of a configuration. Some objects, such as cars, or uk telegram data technical equipment offer complex configurations with rules of precedence, exclusion, etc. In short, a nice puzzle that gets even more complicated when new options appear. A BT, by its modular nature, can be a solution to avoid having to manage an explosive combination.
Whenever we talk about money, we often talk about rules: rules for calculating rates, for combining certain options but not others…
And when we talk about rules, it seems preferable to opt for an approach that is itself based on rules! When their number is high and their intricacy makes a procedural approach risky, then the adaptive approach (BT) offers modularity that favors the evolution, addition and modification of rules.