In modern industrial collaborative robotic applications, it is desirable to create robot programs automatically, intuitively, and time-efficiently. Moreover, robots need to be controlled by reactive policies to face the unpredictability of the environment they operate in. In this paper we propose a framework that combines a method that learns Behavior Trees (BTs) from demonstration with a method that evolves them with Genetic Programming (GP) for collaborative robotic applications. The main contribution of this paper is to show that by combining the two learning methods we obtain a method that allows non-expert users to semi-automatically, time-efficiently, and interactively generate BTs. We validate the framework with a series of manipulation experiments. The BT is fully learnt in simulation and then transferred to a real collaborative robot.

This thesis aims to address the challenge of generating task plans for robots in industry-relevant scenarios. With the increase in small-batch production, companies require robots to be reprogrammed frequently for new tasks. However, maintaining a team of operators with specific programming skills is only cost-efficient for large-scale production. The increase in automation targets companies where humans share their working environment with robots, expanding the scope of manufacturing applications. To achieve that, robots need to be controlled by task plans, which sequence and optimize the execution of actions. This thesis focuses on generating task plans that are reactive, transparent and explainable, modular, and automatically synthesized. These task plans improve the robot’s autonomy, fault-tolerance, and robustness. Furthermore, such task plans facilitate the collaboration with humans, enabling intuitive representations of the plan and the possibility for humans to prompt instructions at run-time to modify the robot’s behavior. Lastly, autonomous generation decreases the programming skills required for the operator to program a robot, and optimizes the task plan. This thesis discusses the use of Behavior Trees (BTs) as policy representations for robotic task plans. It compares the modularity of BTs and Finite State Machines (FSMs) and concludes that BTs are more effective for industrial scenarios. This thesis also explores the automatic and intuitive generation of BTs using Genetic Programming and Learning from Demonstration methods, respectively. The proposed methods aim to time-efficiently evolve BTs for mobile manipulation tasks and allow non-expert users to intuitively teach robots manipulation tasks. This thesis highlights the importance of user experience in task solving and how it can benefit evolutionary algorithms. Finally, it proposes the use of previously learned BTs from demonstration to intervene in the unsupervised learning process.

Den här avhandlingen syftar till att ta itu med utmaningen att generera uppgiftsplaner för robotar i industriella scenarier. Med ökningen av småskalig produktion kräver företag att robotar omprogrammeras frekvent för nya uppgifter. Att upprätthålla en grupp operatörer med specifika programmeringsfärdigheter är dock endast kostnadseffektivt för storskalig produktion. Ökningen av automation riktar sig till företag där människor delar sin arbetsmiljö med robotar och utökar omfattningen av tillverkningsapplikationer. För att uppnå detta måste robotar styras av uppgiftsplaner som sekvenserar och optimerar utförandet av åtgärder. Denna avhandling fokuserar på att generera uppgiftsplaner som är reaktiva, transparenta och förklarbara, modulära och automatiskt syntetiserade. Dessa uppgiftsplaner förbättrar robotens autonomi, feltolerans och robusthet. Dessutom underlättar sådana uppgiftsplaner samarbetet med människor genom att möjliggöra intuitiva representationer av planen och möjligheten för människor att ge instruktioner vid körningstid för att ändra robotens beteende. Slutligen minskar autonom generering programmeringsfärdigheterna som krävs för att operatören ska kunna programmera en robot och optimerar uppgiftsplanen. Denna avhandling diskuterar användningen av beteendeträd (BTs) som policyrepresentationer för robotiska uppgiftsplaner. Den jämför moduleringen av BT och deterministiska tillståndsmaskiner (FSMs) och drar slutsatsen att BTs är mer effektiva för industriella scenarier. Denna avhandling utforskar också den automatiska och intuitiva generationen av BTs med hjälp av genetisk programmering och lärande från demonstrationsmetoder, respektive. De föreslagna metoderna syftar till att tidsmässigt utveckla BTs för mobila manipulationuppgifter och tillåta icke-experter att intuitivt lära robotar manipulationsuppgifter. Denna avhandling belyser vikten av användarupplevelsen i uppgiftslösning och hur den kan gynna evolutionära algoritmer. Slutligen föreslår den användningen av tidigare inlärda BTs från demonstration för att ingripa i den oövervakade inlärningsprocessen.

In this paper we provide a practical demonstration of how the modularity in a Behavior Tree (BT) decreases the effort in programming a robot task when compared to a Finite State Machine (FSM). In recent years the way to represent a task plan to control an autonomous agent has been shifting from the standard FSM towards BTs. Many works in the literature have highlighted and proven the benefits of such design compared to standard approaches, especially in terms of modularity, reactivity and human readability. However, these works have often failed in providing a tangible comparison in the implementation of those policies and the programming effort required to modify them. This is a relevant aspect in many robotic applications, where the design choice is dictated both by the robustness of the policy and by the time required to program it. In this work, we compare backward chained BTs with a fault-tolerant design of FSMs by evaluating the cost to modify them. We validate the analysis with a set of experiments in a simulation environment where a mobile manipulator solves an item fetching task.

Behavior Trees (BTs) were invented as a tool to enable modular AI in computer games, but have received an increasing amount of attention in the robotics community in the last decade. With rising demands on agent AI complexity, game programmers found that the Finite State Machines (FSM) that they used scaled poorly and were difficult to extend, adapt and reuse. In BTs, the state transition logic is not dispersed across the individual states, but organized in a hierarchical tree structure, with the states as leaves. This has a significant effect on modularity, which in turn simplifies both synthesis and analysis by humans and algorithms alike. These advantages are needed not only in game AI design, but also in robotics, as is evident from the research being done. In this paper we present a comprehensive survey of the topic of BTs in Artificial Intelligence and Robotic applications. The existing literature is described and categorized based on methods, application areas and contributions, and the paper is concluded with a list of open research challenges.

Fast changing tasks in unpredictable, collaborative environments are typical for medium-small companies, where robotised applications are increasing. Thus, robot programs should be generated in short time with small effort, and the robot able to react dynamically to the environment. To address this we propose a method that combines context awareness and planning to learn Behavior Trees (BTs), a reactive policy representation that is becoming more popular in robotics and has been used successfully in many collaborative scenarios. Context awareness allows for inferring from the demonstration the frames in which actions are executed and to capture relevant aspects of the task, while a planner is used to automatically generate the BT from the sequence of actions from the demonstration. The learned BT is shown to solve non-trivial manipulation tasks where learning the context is fundamental to achieve the goal. Moreover, we collected non-expert demonstrations to study the performances of the algorithm in industrial scenarios.

Industrial robots can solve tasks in controlled environments, but modern applications require robots able to operate also in unpredictable surroundings. An increasingly popular reactive policy architecture in robotics is Behavior Trees (BTs) but as other architectures, programming time drives cost and limits flexibility. The two main branches of algorithms to generate policies automatically, automated planning and machine learning, both have their own drawbacks and have not previously been combined for generation of BTs. We propose a method for creating BTs by combining these branches, inserting the result of an automated planner into the population of a Genetic Programming algorithm. Experiments confirm that the proposed method performs well on a variety of robotic assembly problems and outperforms the base methods used separately. We also show that this high level learning of Behavior Trees can be transferred to a real system without further training. 

Abstract:In recent years, robots are used in an increasing variety of tasks, especially by small- and medium sized enterprises. These tasks are usually fast-changing, they have a collaborative scenario and happen in unpredictable environments with possible ambiguities. It is important to have methods capable of generating robot programs easily, that are made as general as possible by handling uncertainties. We present a system that integrates a method to learn Behavior Trees (BTs) from demonstration for pick and place tasks, with a framework that uses verbal interaction to ask follow-up clarification questions to resolve ambiguities. During the execution of a task, the system asks for user input when there is need to disambiguate an object in the scene, i.e. when the targets of the task are objects of a same type that are present in multiple instances. The integrated system is demonstrated on different scenarios of a pick and place task, with increasing level of ambiguities. The code used for this paper is made publicly available 1 1 https://github.com/matiov/disambiguate-BT-execution.

Modern industrial applications require robots to operate in unpredictable environments, and programs to be created with a minimal effort, to accommodate frequent changes to the task. Here, we show that genetic programming can be effectively used to learn the structure of a behavior tree (BT) to solve a robotic task in an unpredictable environment. We propose to use a simple simulator for learning, and demonstrate that the learned BTs can solve the same task in a realistic simulator, converging without the need for task specific heuristics, making our method appealing for real robotic applications.

In modern industrial collaborative robotic applications, it is desirable to create robot programs automatically,intuitively, and time-efficiently. Moreover, robots need to becontrolled by reactive policies to face the unpredictability ofthe environment they operate in. In this paper we proposea framework that combines a method that learns BehaviorTrees (BTs) from demonstration with a method that evolvesthem with Genetic Programming (GP) for collaborative roboticapplications. The main contribution of this paper is to show thatby combining the two learning methods we obtain a method thatallows non-expert users to semi-automatically, time-efficiently,and interactively generate BTs. We validate the framework witha series of manipulation experiments. The BT is fully learnt insimulation and then transferred to a real collaborative robot.