Reinforcement Learning (RL) has emerged as a transformative approach in the domains of automation and robotics, offering powerful solutions to complex problems that conventional methods struggle to address. In scenarios where the problem definitions are elusive and challenging to quantify, learning-based solutions such as RL become particularly valuable. One instance of such complexity can be found in the realm of car racing, a dynamic and unpredictable environment that demands sophisticated decision-making algorithms. This study focuses on developing and training an RL agent to navigate a racing environment solely using feedforward raw lidar and velocity data in a simulated context. The agent's performance, trained in the simulation environment, is then experimentally evaluated in a real-world racing scenario. This exploration underlines the feasibility and potential benefits of RL algorithm enhancing autonomous racing performance, especially in the environments where prior map information is not available.

Deep Reinforcement Learning (DRL) enables cognitive Autonomous Ground Vehicle (AGV) navigation utilizing raw sensor data without a-priori maps or GPS, which is a necessity in hazardous, information poor environments such as regions where natural disasters occur, and extraterrestrial planets. The substantial training time required to learn an optimal DRL policy, which can be days or weeks for complex tasks, is a major hurdle to real-world implementation in AGV applications. Training entails repeated collisions with the surrounding environment over an extended time period, dependent on the complexity of the task, to reinforce positive exploratory, application specific behavior that is expensive, and time consuming in the real-world. Effectively bridging the simulation to real-world gap is a requisite for successful implementation of DRL in complex AGV applications, enabling learning of cost-effective policies. We present AutoVRL, an open-source high fidelity simulator built upon the Bullet physics engine utilizing OpenAI Gym and Stable Baselines3 in PyTorch to train AGV DRL agents for sim-to-real policy transfer. AutoVRL is equipped with sensor implementations of GPS, IMU, LiDAR and camera, actuators for AGV control, and realistic environments, with extensibility for new environments and AGV models. The simulator provides access to state-of-the-art DRL algorithms, utilizing a python interface for simple algorithm and environment customization, and simulation execution.

Autonomous robots would benefit a lot by gaining the ability to manipulate their environment to solve path planning tasks, known as the Navigation Among Movable Obstacle (NAMO) problem. In this paper, we present a deep reinforcement learning approach for solving NAMO locally, near narrow passages. We train parallel agents in physics simulation using an Advantage Actor-Critic based algorithm with a multi-modal neural network. We present an online policy that is able to push obstacles in a non-axial-aligned fashion, react to unexpected obstacle dynamics in real-time, and solve the local NAMO problem. Experimental validation in simulation shows that the presented approach generalises to unseen NAMO problems in unknown environments. We further demonstrate the implementation of the policy on a real quadrupedal robot, showing that the policy can deal with real-world sensor noises and uncertainties in unseen NAMO tasks.

An agent's ability to leverage past experience is critical for efficiently solving new tasks. Approximate solutions for new tasks can be obtained from previously derived value functions, as demonstrated by research on transfer learning, curriculum learning, and compositionality. However, prior work has primarily focused on using value functions to obtain zero-shot approximations for solutions to a new task. In this work, we show how an arbitrary approximation for the value function can be used to derive double-sided bounds on the optimal value function of interest. We further extend the framework with error analysis for continuous state and action spaces. The derived results lead to new approaches for clipping during training which we validate numerically in simple domains.

We consider the problem of optimal unsignalized intersection management for continual streams of randomly arriving robots. This problem involves repeatedly solving different instances of a mixed integer program, for which the computation time using a naive optimization algorithm scales exponentially with the number of robots and lanes. Hence, such an approach is not suitable for real-time implementation. In this paper, we propose a solution framework that combines learning and sequential optimization. In particular, we propose an algorithm for learning a shared policy that given the traffic state information, determines the crossing order of the robots. Then, we optimize the trajectories of the robots sequentially according to that crossing order. This approach inherently guarantees safety at all times. We validate the performance of this approach using extensive simulations. Our approach, on average, significantly outperforms the heuristics from the literature. We also show through simulations that the computation time for our approach scales linearly with the number of robots. We further implement the learnt policies on physical robots with a few modifications to the solution framework to address real-world challenges and establish its real-time implementability.

Contextual bandit is a widely used model for sequential decision making. The interaction between the learner and the environment proceeds in rounds: in each round, the environment provides a context; based on it, the learner chooses an action and receive a reward. The goal is to maximize the total reward across multiple rounds. This model has found extensive applications in fields such as medical treatment [Tewari and Murphy, 2017], personalized recommendations [Beygelzimer et al., 2011], and online advertising [Chu et al., 2011]. Algorithms for contextual bandits with provable guarantees have been developed under various assumptions. In the linear regime, the most extensively studied model is the stochastic linear contextual bandit, in which the context can be arbitrarily distributed in each round, while the reward is determined by a fixed linear function of the context-action pair. Near-optimal algorithms for this setting have been established in, e.g., [Chu et al., 2011, Abbasi-Yadkori et al., 2011, Li et al., 2019, Foster et al., 2020]. Another model, which is the focus of this paper, is the adversarial linear contextual bandit, in which the context is drawn from a fixed distribution, while the reward is determined by a time-varying linear function of the context-action pair.

In dynamic motion generation tasks, including contact and collisions, small changes in policy parameters can lead to extremely different returns. For example, in soccer, the ball can fly in completely different directions with a similar heading motion by slightly changing the hitting position or the force applied to the ball or when the friction of the ball varies. However, it is difficult to imagine that completely different skills are needed for heading a ball in different directions. In this study, we proposed a multitask reinforcement learning algorithm for adapting a policy to implicit changes in goals or environments in a single motion category with different reward functions or physical parameters of the environment. We evaluated the proposed method on the ball heading task using a monopod robot model. The results showed that the proposed method can adapt to implicit changes in the goal positions or the coefficients of restitution of the ball, whereas the standard domain randomization approach cannot cope with different task settings.

Due to the proliferation of malware, defenders are increasingly turning to automation and machine learning as part of the malware detection tool-chain. However, machine learning models are susceptible to adversarial attacks, requiring the testing of model and product robustness. Meanwhile, attackers also seek to automate malware generation and evasion of antivirus systems, and defenders try to gain insight into their methods. This work proposes a new algorithm that combines Malware Evasion and Model Extraction (MEME) attacks. MEME uses model-based reinforcement learning to adversarially modify Windows executable binary samples while simultaneously training a surrogate model with a high agreement with the target model to evade. To evaluate this method, we compare it with two state-of-the-art attacks in adversarial malware creation, using three well-known published models and one antivirus product as targets. Results show that MEME outperforms the state-of-the-art methods in terms of evasion capabilities in almost all cases, producing evasive malware with an evasion rate in the range of 32-73%. It also produces surrogate models with a prediction label agreement with the respective target models between 97-99%. The surrogate could be used to fine-tune and improve the evasion rate in the future.

A Markov decision process can be parameterized by a transition kernel and a reward function. Both play essential roles in the study of reinforcement learning as evidenced by their presence in the Bellman equations. In our inquiry of various kinds of "costs" associated with reinforcement learning inspired by the demands in robotic applications, rewards are central to understanding the structure of a Markov decision process and reward-centric notions can elucidate important concepts in reinforcement learning. Specifically, we study the sample complexity of policy evaluation and develop a novel estimator with an instance-specific error bound of $\tilde{O}(\sqrt{\frac{\tau_s}{n}})$ for estimating a single state value. Under the online regret minimization setting, we refine the transition-based MDP constant, diameter, into a reward-based constant, maximum expected hitting cost, and with it, provide a theoretical explanation for how a well-known technique, potential-based reward shaping, could accelerate learning with expert knowledge. In an attempt to study safe reinforcement learning, we model hazardous environments with irrecoverability and proposed a quantitative notion of safe learning via reset efficiency. In this setting, we modify a classic algorithm to account for resets achieving promising preliminary numerical results. Lastly, for MDPs with multiple reward functions, we develop a planning algorithm that computationally efficiently finds Pareto-optimal stochastic policies.

Dialogue policy learning (DPL) is a crucial component of dialogue modelling. Its primary role is to determine the appropriate abstract response, commonly referred to as the "dialogue action". Traditional DPL methodologies have treated this as a sequential decision problem, using pre-defined action candidates extracted from a corpus. However, these incomplete candidates can significantly limit the diversity of responses and pose challenges when dealing with edge cases, which are scenarios that occur only at extreme operating parameters. To address these limitations, we introduce a novel framework, JoTR. This framework is unique as it leverages a text-to-text Transformer-based model to generate flexible dialogue actions. Unlike traditional methods, JoTR formulates a word-level policy that allows for a more dynamic and adaptable dialogue action generation, without the need for any action templates. This setting enhances the diversity of responses and improves the system's ability to handle edge cases effectively. In addition, JoTR employs reinforcement learning with a reward-shaping mechanism to efficiently finetune the word-level dialogue policy, which allows the model to learn from its interactions, improving its performance over time. We conducted an extensive evaluation of JoTR to assess its effectiveness. Our extensive evaluation shows that JoTR achieves state-of-the-art performance on two benchmark dialogue modelling tasks, as assessed by both user simulators and human evaluators.