Real-world sequential decision-making tasks are usually complex, and require trade-offs between multiple–often conflicting–objectives. However, the majority of research in reinforcement learning (RL) and decision-theoretic planning assumes a single objective, or that multiple objectives can be handled via a predefined weighted sum over the objectives. Such approaches may oversimplify the underlying problem, and produce suboptimal results. This extended abstract outlines the limitations of using a semi-blind iterative process to solve multi-objective decision making problems. Our extended paper [4], serves as a guide for the application of explicitly multi-objective methods to difficult problems.

Many real-world problems require a trade-off between multiple conflicting objectives. Decision-makers’ preferences over solutions to such problems are determined by their utility functions, which convert multi-objective values to scalars. In some settings, utility functions change over time, and the goal is to find methods that can efficiently adapt an agent’s policy to changes in utility. Previous work on learning with dynamic utility functions has focused on model-free methods, which often suffer from poor sample efficiency. In this work, we instead propose a model-based actor-critic, which explores with diverse utility functions through imagined rollouts within a learned world model between interactions with the real environment. An experimental evaluation on Minecart, a well-known benchmark for multi-objective reinforcement learning, shows that by learning a model of the environment the quality of the agent’s policy is improved compared to model-free algorithms.

Many real-world problems require a trade-off between multiple conflicting objectives. Decision-makers’ preferences over solutions to such problems are determined by their utility functions, which convert multi-objective values to scalars. In some settings, utility functions change over time, and the goal is to find methods that can efficiently adapt an agent’s policy to changes in utility. Previous work on learning with dynamic utility functions has focused on model-free methods, which often suffer from poor sample efficiency. In this work, we instead propose a model-based actor-critic, which explores with diverse utility functions through imagined rollouts within a learned world model between interactions with the real environment. An experimental evaluation shows that by learning a model of the environment the performance of the agent can be improved compared to model-free algorithms.

Team training in complex domains often requires a substantial number of resources, e.g. vehicles, machines, and role-players. For this reason, it may be difficult to realise efficient and effective training scenarios in a real-world setting. Instead, part of the training can be conducted in synthetic, computer-generated environments. In these environments trainees can operate simulators instead of real vehicles, while synthetic actors can replace human role-players to increase the complexity of the simulated scenario at low operating cost. However, constructing behaviour models for synthetic actors is challenging, especially for the end users, who typically do not have expertise in artificial intelligence. In this dissertation, we study how machine learning can be used to simplify the construction of intelligent agents for simulation-based training. A simulation-based air combat training system is used as case study. 

The contributions of the dissertation are divided into two parts. The first part aims at improving the understanding of reinforcement learning in the domain of simulation-based training. First, a user-study is conducted to identify important capabilities and characteristics of learning agents that are intended to support training of fighter pilots. It is identified that one of the most important capabilities of learning agents in the context of simulation-based training is that their behaviour can be adapted to different phases of training, as well as to the training needs of individual human trainees. Second, methods for learning how to coordinate with other agents are studied in simplified training scenarios, to investigate how the design of the agent’s observation space, action space, and reward signal affects the performance of learning. It is identified that temporal abstractions and hierarchical reinforcement learning can improve the efficiency of learning, while also providing support for modelling of doctrinal behaviour. In more complex settings, curriculum learning and related methods are expected to help find novel tactics even when sparse, abstract reward signals are used. Third, based on the results from the user study and the practical experiments, a system concept for a user-adaptive training system is developed to support further research. 

The second part of the contributions focuses on methods for utility-based multi-objective reinforcement learning, which incorporates knowledge of the user’s utility function in the search for policies that balance multiple conflicting objectives. Two new agents for multi-objective reinforcement learning are proposed: the Tunable Actor (T-Actor) and the Multi-Objective Dreamer (MO-Dreamer). T-Actor provides decision support to instructors by learning a set of Pareto optimal policies, represented by a single neural network conditioned on objective preferences. This enables tuning of the agent’s behaviour to fit trainees’ current training needs. Experimental evaluations in gridworlds and in the target system show that T-Actor reduces the number of training steps required for learning. MO-Dreamer adapts online to changes in users’ utility, e.g. changes in training needs. It does so by learning a model of the environment, which it can use for anticipatory rollouts with a diverse set of utility functions to explore which policy to follow to optimise the return for a given set of objective preferences. An experimental evaluation shows that MO-Dreamer outperforms prior model-free approaches in terms of experienced regret, for frequent as well as sparse changes in utility. 

Overall, the research conducted in this dissertation contributes to improved knowledge about how to apply machine learning methods to construction of simulation-based training environments. While our focus was on air combat training, the results are general enough to be applicable in other domains. 

Silver et al.[14] posit that scalar reward maximisation is sufficient to underpin all intelligence and provides a suitable basis for artificial general intelligence (AGI). This extended abstract summarises the counter-argument from our JAAMAS paper [19].

Real-world sequential decision-making tasks are generally complex, requiring trade-offs between multiple, often conflicting, objectives. Despite this, the majority of research in reinforcement learning and decision-theoretic planning either assumes only a single objective, or that multiple objectives can be adequately handled via a simple linear combination. Such approaches may oversimplify the underlying problem and hence produce suboptimal results. This paper serves as a guide to the application of multi-objective methods to difficult problems, and is aimed at researchers who are already familiar with single-objective reinforcement learning and planning methods who wish to adopt a multi-objective perspective on their research, as well as practitioners who encounter multi-objective decision problems in practice. It identifies the factors that may influence the nature of the desired solution, and illustrates by example how these influence the design of multi-objective decision-making systems for complex problems.

The high operational cost of aircraft, limited availability of air space, and strict safety regulations make training of fighter pilots increasingly challenging. By integrating Live, Virtual, and Constructive simulation resources, efficiency and effectiveness can be improved. In particular, if constructive simulations, which provide synthetic agents operating synthetic vehicles, were used to a higher degree, complex training scenarios could be realised at low cost, the need for support personnel could be reduced, and training availability could be improved. In this work, inspired by the recent improvements of techniques for artificial intelligence, we take a user perspective and investigate how intelligent, learning agents could help build future training systems. Through a domain analysis, a user study, and practical experiments, we identify important agent capabilities and characteristics, and then discuss design approaches and solution concepts for training systems to utilise learning agents for improved training value.

The recent paper “Reward is Enough” by Silver, Singh, Precup and Sutton posits that the concept of reward maximisation is sufficient to underpin all intelligence, both natural and artificial, and provides a suitable basis for the creation of artificial general intelligence. We contest the underlying assumption of Silver et al. that such reward can be scalar-valued. In this paper we explain why scalar rewards are insufficient to account for some aspects of both biological and computational intelligence, and argue in favour of explicitly multi-objective models of reward maximisation. Furthermore, we contend that even if scalar reward functions can trigger intelligent behaviour in specific cases, this type of reward is insufficient for the development of human-aligned artificial general intelligence due to unacceptable risks of unsafe or unethical behaviour.

The high operational cost of aircraft, limited availability of air space, and strict safety regulations make training of fighter pilots increasingly challenging. By integrating Live, Virtual, and Constructive simulation resources, efficiency and effectiveness can be improved. In particular, if constructive simulations, which provide synthetic agents operating synthetic vehicles, were used to a higher degree, complex training scenarios could be realized at low cost, the need for support personnel could be reduced, and training availability could be improved. In this work, inspired by the recent improvements of techniques for artificial intelligence, we take a user perspective and investigate how intelligent, learning agents could help build future training systems. Through a domain analysis, a user study, and practical experiments, we identify important agent capabilities and characteristics, and then discuss design approaches and solution concepts for training systems to utilize learning agents for improved training value.

Agent-based simulation can be used for efficient and effective training of human operators and decision-makers. However, constructing realistic behavior models for the agents is challenging and time-consuming, especially for subject matter experts, who may not have expertise in artificial intelligence. In this work, we investigate how machine learning can be used to adapt simulation contents to the current needs of individual trainees. Our initial results demonstrate that multi-objective multi-agent reinforcement learning is a promising approach for creating agents with diverse and adaptive characteristics, which can stimulate humans in training.