Model-free reinforcement learning for motion planning of autonomous agents with complex tasks in partially observable environments

Swarm systems can cooperatively and efficiently accomplish specified complex tasks. Recent works have shown the potential of multi-agent reinforcement learning methods to study behavior policies of swarm systems. However, it is difficult for them to complete complex swarm tasks efficiently. In human society, role assignment can effectively help humans understand complex tasks and decompose them into simple certain subtasks. Inspired by this, we propose a two-level hierarchical cooperative swarm policy learning framework with role emergence based on hierarchical deep reinforcement learning for distributed swarm systems. In this framework, roles are dynamic and emergent. Agents with the same role tend to collectively complete a certain subtask. Specifically, each agent uses a higher-level swarm policy to dynamically select a role for itself in a role space and at a higher temporal scale, while it uses a lower-level swarm policy to perform the responsibilities of the selected role in a primitive action space. Meanwhile, hierarchical swarm policies with partial observation are centrally trained and decentrally executed, where agents' local interaction modules and extrinsic team rewards are designed to promote cooperation among agents. In addition, an intrinsic reward is defined to enable different roles to be identified by agents' longer-term behaviors, which implicitly associates the roles with responsibilities. Simulation results show that our method can learn and generate emergent, dynamic and identifiable roles, which helps swarm systems to reliably and efficiently accomplish complex tasks in a shorter time.

Retrieving objects from clutters is a complex task, which requires multiple interactions with the environment until the target object can be extracted. These interactions involve executing action primitives like grasping or pushing as well as setting priorities for the objects to manipulate and the actions to execute. Mechanical Search (MS) [1] is a framework for object retrieval, which uses a heuristic algorithm for pushing and rule-based algorithms for high-level planning. While rule-based policies profit from human intuition in how they work, they usually perform sub-optimally in many cases. Deep reinforcement learning (RL) has shown great performance in complex tasks such as taking decisions through evaluating pixels, which makes it suitable for training policies in the context of object-retrieval. In this work, we first formulate the MS problem in a principled formulation as a hierarchical POMDP. Based on this formulation, we propose a hierarchical policy learning approach for the MS problem. For demonstration, we present two main parameterized sub-policies: a push policy and an action selection policy. When integrated into the hierarchical POMDP's policy, our proposed sub-policies increase the success rate of retrieving the target object from less than 32% to nearly 80%, while reducing the computation time for push actions from multiple seconds to less than 10 milliseconds.

Reinforcement Learning (RL) aims at learning an optimal behavior policy from its own experiments and not rule-based control methods. However, there is no RL algorithm yet capable of handling a task as difficult as urban driving. We present a novel technique, coined implicit affordances, to effectively leverage RL for urban driving thus including lane keeping, pedestrians and vehicles avoidance, and traffic light detection. To our knowledge we are the first to present a successful RL agent handling such a complex task especially regarding the traffic light detection. Furthermore, we have demonstrated the effectiveness of our method by winning the Camera Only track of the CARLA challenge.

To behave adaptively, we must learn from the consequences of our actions. Doing so is difficult when the consequences of an action follow a delay. This introduces the problem of temporal credit assignment. When feedback follows a sequence of decisions, how should the individual assign credit to the intermediate actions that comprise the sequence? Research in reinforcement learning provides 2 general solutions to this problem: model-free reinforcement learning and model-based reinforcement learning. In this review, we examine connections between stimulus-response and cognitive learning theories, habitual and goal-directed control, and model-free and model-based reinforcement learning. We then consider a range of problems related to temporal credit assignment. These include second-order conditioning and secondary reinforcers, latent learning and detour behavior, partially observable Markov decision processes, actions with distributed outcomes, and hierarchical learning. We ask whether humans and animals, when faced with these problems, behave in a manner consistent with reinforcement learning techniques. Throughout, we seek to identify neural substrates of model-free and model-based reinforcement learning. The former class of techniques is understood in terms of the neurotransmitter dopamine and its effects in the basal ganglia. The latter is understood in terms of a distributed network of regions including the prefrontal cortex, medial temporal lobes, cerebellum, and basal ganglia. Not only do reinforcement learning techniques have a natural interpretation in terms of human and animal behavior but they also provide a useful framework for understanding neural reward valuation and action selection.

Reinforcement learning promises high performance in complex tasks as well as low online storage and computation cost. However, the trial-and-error learning approach of reinforcement learning could explore unsafe behavior in the search for an optimal solution. Run time assurance (RTA) approaches can be applied to monitor behavior and ensure safety constraint satisfaction during reinforcement learning. This paper investigates the effect of RTA on reinforcement learning training performance in terms of training efficiency, safety constraint satisfaction, control efficiency, task efficiency, and training duration. For the purposes of demonstration, a custom reinforcement learning environment is created where the objective is to develop a policy that moves a satellite into docking position with another satellite in a two-dimensional relative-motion reference frame. Six different policies are trained. The first features no RTA, the second features no RTA but a higher penalty for safety violations, and four others use different RTA techniques to enforce a dynamic velocity constraint during training. The trained policies are analyzed with standardized test points. It is shown that the policies trained without RTA do not learn to adhere to the constraint, whereas all policies trained with RTA do learn to adhere to the constraint. Although more complex RTA frameworks can be better for operational use, it is found that a simple RTA framework provides the best overall results for reinforcement learning training.

Deep reinforcement learning (RL) is achieving significant success in various applications like control, robotics, games, resource management, and scheduling. However, the important problem of emergency evacuation, which clearly could benefit from RL, has been largely unaddressed. Indeed, emergency evacuation is a complex task that is difficult to solve with RL. An emergency situation is highly dynamic, with a lot of changing variables and complex constraints that make it challenging to solve. Also, there is no standard benchmark environment available that can be used to train RL agents for evacuation. A realistic environment can be complex to design. In this article, we propose the first fire evacuation environment to train RL agents for evacuation planning. The environment is modeled as a graph capturing the building structure. It consists of realistic features like fire spread, uncertainty, and bottlenecks. The implementation of our environment is in the OpenAI gym format, to facilitate future research. We also propose a new RL approach that entails pretraining the network weights of a DQN-based agent [DQN/Double-DQN (DDQN)/Dueling-DQN] to incorporate information on the shortest path to the exit. We achieved this by using tabular <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$Q$ </tex-math></inline-formula> -learning to learn the shortest path on the building model’s graph. This information is transferred to the network by deliberately overfitting it on the <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$Q$ </tex-math></inline-formula> -matrix. Then, the pretrained DQN model is trained on the fire evacuation environment to generate the optimal evacuation path under time varying conditions due to fire spread, bottlenecks, and uncertainty. We perform comparisons of the proposed approach with state-of-the-art RL algorithms like DQN, DDQN, Dueling-DQN, PPO, VPG, state-action-reward-state-action (SARSA), actor–critic method, and ACKTR. The results show that our method is able to outperform state-of-the-art models by a huge margin including the original DQN-based models. Finally, our model is tested on a large and complex real building consisting of 91 rooms, with the possibility to move to any other room, hence giving 8281 actions. In order to reduce the action space, we propose a strategy that involves one step simulation. That is, an action importance vector is added to the final output of the pretrained DQN and acts like an attention mechanism. Using this strategy, the action space is reduced by 90.1%. In this manner, the model is able to deal with large action spaces. Hence, our model achieves near optimal performance on the real world emergency environment.

Advancements in reinforcement learning (RL) have demonstrated superhuman performance in complex tasks such as Starcraft, Go, Chess etc. However, knowledge transfer from Artificial ``Experts" to humans remain a significant challenge. A promising avenue for such transfer would be the use of curricula. Recent methods in curricula generation focuses on training RL agents efficiently, yet such methods rely on surrogate measures to track student progress, and are not suited for training robots in the real world (or more ambitiously humans). In this paper, we introduce a method named Parameterized Environment Response Model (PERM) that shows promising results in training RL agents in parameterized environments. Inspired by Item Response Theory, PERM seeks to model difficulty of environments and ability of RL agents directly. Given that RL agents and humans are trained more efficiently under the ``zone of proximal development", our method generates a curriculum by matching the difficulty of an environment to the current ability of the student. In addition, PERM can be trained offline and does not employ non-stationary measures of student ability, making it suitable for transfer between students. We demonstrate PERM's ability to represent the environment parameter space, and training with RL agents with PERM produces a strong performance in deterministic environments. Lastly, we show that our method is transferable between students, without any sacrifice in training quality.

Technological advancements in robotics and cognitive science are contributing to the development of the field of cognitive robotics. Modern robotic platforms are able to exhibit the ability to learn and reason about complex tasks and to follow behavioural goals in complex environments. Nevertheless, many challenges still exist. One of these great challenges is to equip these robots with cognitive systems that allow them to deal with less constrained situations, beyond constrained scenarios as in industrial robotics. In this work we explore the application of the Reinforcement Learning (RL) paradigm to study the autonomous development of robot controllers without a priori supervised learning. Such a model-based RL architecture is discussed for the cognitive implications of applying RL in humanoid robots. To this end we show a developmental framework for RL in robotics and its implementation and testing for the iCub robotic platform in two novel experimental scenarios. In particular we focus on iCub simulation experiments with comparisons between internal perception-based reward signals and external ones, in order to compare learning performance of the robot guided by its own perception of action's outcomes with the one when the robot has its actions externally evaluated.

Due to its wide application, deep reinforcement learning (DRL) has been extensively studied in the motion planning community in recent years. However, in the current DRL research, regardless of task completion, the state information of the agent will be reset afterward. This leads to a low sample utilization rate and hinders further explorations of the environment. Moreover, in the initial training stage, the agent has a weak learning ability in general, which affects the training efficiency in complex tasks. In this study, a new hierarchical reinforcement learning (HRL) framework dubbed hierarchical learning based on game playing with state relay (HGR) is proposed. In particular, we introduce an auxiliary penalty to regulate task difficulty, and one training mechanism, the state relay mechanism, is designed. The relay mechanism can make full use of the intermediate states of the agent and expand the environment exploration of low-level policy. Our algorithm can improve the sample utilization rate, reduce the sparse reward problem, and thereby enhance the training performance in complex environments. Simulation tests are carried out on two public experiment platforms, i.e., MazeBase and MuJoCo, to verify the effectiveness of the proposed method. The results show that HGR significantly benefits the reinforcement learning (RL) area.

In video established eye tracking methods, there are both mechanical and electrical based approaches existing. With the emerging spread of gaze tracking technology in the recent years and its significance in daily life routine, the data content acquired from the eye behavior tracing turn into important. Several research works were proposed to track the behavior of gaze while playing videos. Tracking an eye gaze while playing a dynamic videos consisting of numerous frames is a complex problem which needs excessive computational efforts. To handle such a complex task, this research proposes Reinforcement Learning (RL) based gaze behavior prediction model. These techniques are found to be invasive in nature and for visual attention behavior analysis applications, these invasive eye tracking system is not applicable. Hence the non-invasive eye tracking could be developed by determining the point of gaze based on observed image processing techniques. Some of the prevailing techniques include artificial intelligence, deep learning, and reinforcement learning and so on. Though quite a few research works has been admitted in this research area, there are several challenges existing so far. The suggested learning techniques are found to be computationally complex and time consuming. This current research work intends to propose a deep convolutional reinforcement learning (DC-RL) model for predicting the visual attention behavior of a person over dynamic scenes.

Reinforcement Learning (RL) algorithms depend on well-defined reward functions for policy optimization. Designing such functions is a complex task, even for domain experts. However, valid task demonstrations can still be collected from moderately skilled individuals. This motivates the use of methods such as Imitation Learning (IL) and Inverse Reinforcement Learning (IRL), where an expert provides demonstrations, allowing the algorithm to infer a policy or reward function that aligns with observed behavior. A common assumption in IRL is that demonstrations come from highly skilled experts. While some studies have explored the impact of suboptimal demonstrators, the influence of intentionally malicious demonstrations remains underexplored. This study introduces an adversarial demonstrator framework that systematically perturbs a subset of demonstrations to manipulate the reward function inferred by an IRL algorithm. Additionally, it quantifies the impact of such adversarial manipulations on the learned policy. Our results show that simply altering 10% of the demonstrations can lead the IRL algorithm to learn a faulty reward function, ultimately degrading the performance of the trained policy by up to three times the effect of adding 10% random trajectories to the result.

We present a two-layered architecture for task-agent assignment and motion planning of Bi-manual Mobile manipulators for executing complex tasks. We use search trees in temporal windows to determine feasible task assignments of agents using task and spatial constraint-based heuristics. We also introduce a caching scheme for moving between different trees so as to avoid re-planning of those portions of the robot motion that were successful. This greatly reduces the number of calls to the motion planner as compared to direct motion planning after task-agent assignment. We have shown our approach works on a set of complex tasks with significantly lower computation times.

Adapting the user interface (UI) of software systems to meet the needs and preferences of users is a complex task. The main challenge is to provide the appropriate adaptations at the appropriate time to offer value to end-users. Recent advances in Machine Learning (ML) techniques may provide effective means to support the adaptation process. In this paper, we instantiate a reference framework for Intelligent User Interface Adaptation by using Reinforcement Learning (RL) as the ML component to adapt user interfaces and ultimately improving the overall User Experience (UX). By using RL, the system is able to learn from past adaptations to improve the decision-making capabilities. Moreover, assessing the success of such adaptations remains a challenge. To overcome this issue, we propose to use predictive Human-Computer Interaction (HCI) models to evaluate the outcome of each action (i.e., adaptations) performed by the RL agent. In addition, we present an implementation of the instantiated framework, which is an extension of OpenAI Gym, that serves as a toolkit for developing and comparing RL algorithms. This Gym environment is highly configurable and extensible to other UI adaptation contexts. The evaluation results show that our RL-based framework can successfully train RL agents able to learn how to adapt UIs in a specific context to maximize the user engagement by using an HCI model as rewards predictor.

Testing Deep Learning (DL) systems is a complex task as they do not behave like traditional systems would, notably because of their stochastic nature. Nonetheless, being able to adapt existing testing techniques such as Mutation Testing (MT) to DL settings would greatly improve their potential verifiability. While some efforts have been made to extend MT to the Supervised Learning paradigm, little work has gone into extending it to Reinforcement Learning (RL) which is also an important component of the DL ecosystem but behaves very differently from SL. This paper builds on the existing approach of MT in order to propose a framework, RLMutation, for MT applied to RL. Notably, we use existing taxonomies of faults to build a set of mutation operators relevant to RL and use a simple heuristic to generate test cases for RL. This allows us to compare different mutation killing definitions based on existing approaches, as well as to analyze the behavior of the obtained mutation operators and their potential combinations called Higher Order Mutation(s) (HOM). We show that the design choice of the mutation killing definition can affect whether or not a mutation is killed as well as the generated test cases. Moreover, we found that even with a relatively small number of test cases and operators we manage to generate HOM with interesting properties which can enhance testing capability in RL systems.

Opportunities and challenges in applying reinforcement learning to robotic manipulation: An industrial case study