One of the most intriguing learning methods in machine learning literature is reinforcement learning. Through reward and punishment, an agent learns to reach a given goal by selecting an action from a set of actions in a known or unknown environment. Reinforcement learning, unlike supervised and unsupervised learning techniques, does not require any initial data. In this article, we will demonstrate how to implement a version of the reinforcement learning technique Deep Q-Learning, to create an AI agent capable of playing Checkers at a decent level. Deep reinforcement learning is a branch of machine learning that combines deep learning and reinforcement learning (RL).

By Aleksandras Šulženko Image credit: 123RF (with modifications) Imagine you are about to sit down to play a game with a friend. But this isn’t just any friend - it’s a computer program that doesn’t know the rules of the game. It does, however, understand that it has a goal, and that goal is to…

Deep Reinforcement Learning (DRL) is regarded as a potential method for car-following control and has been mostly studied to support a single following vehicle. However, it is more challenging to learn a stable and efficient car-following policy when there are multiple following vehicles in a platoon, especially with unpredictable leading vehicle behavior. In this context, we adopt an integrated DRL and Dynamic Programming (DP) approach to learn autonomous platoon control policies, which embeds the Deep Deterministic Policy Gradient (DDPG) algorithm into a finite-horizon value iteration framework. Although the DP framework can improve the stability and performance of DDPG, it has the limitations of lower sampling and training efficiency. In this paper, we propose an algorithm, namely Finite-Horizon-DDPG with Sweeping through reduced state space using Stationary approximation (FH-DDPG-SS), which uses three key ideas to overcome the above limitations, i.e., transferring network weights backward in time, stationary policy approximation for earlier time steps, and sweeping through reduced state space. In order to verify the effectiveness of FH-DDPG-SS, simulation using real driving data is performed, where the performance of FH-DDPG-SS is compared with those of the benchmark algorithms. Finally, platoon safety and string stability for FH-DDPG-SS are demonstrated.

Process optimization for metal additive manufacturing (AM) is crucial to ensure repeatability, control microstructure, and minimize defects. Despite efforts to address this via the traditional design of experiments and statistical process mapping, there is limited insight on an on-the-fly optimization framework that can be integrated into a metal AM system. Additionally, most of these methods, being data-intensive, cannot be supported by a metal AM alloy or system due to budget restrictions. To tackle this issue, the article introduces a Reinforcement Learning (RL) methodology transformed into an optimization problem in the realm of metal AM. An off-policy RL framework based on Q-learning is proposed to find optimal laser power ($P$) - scan velocity ($v$) combinations with the objective of maintaining steady-state melt pool depth. For this, an experimentally validated Eagar-Tsai formulation is used to emulate the Laser-Directed Energy Deposition environment, where the laser operates as the agent across the $P-v$ space such that it maximizes rewards for a melt pool depth closer to the optimum. The culmination of the training process yields a Q-table where the state ($P,v$) with the highest Q-value corresponds to the optimized process parameter. The resultant melt pool depths and the mapping of Q-values to the $P-v$ space show congruence with experimental observations. The framework, therefore, provides a model-free approach to learning without any prior.

Given an environment (e.g., a simulator) for evaluating samples in a specified design space and a set of weighted evaluation metrics -- one can use Theta-Resonance, a single-step Markov Decision Process (MDP), to train an intelligent agent producing progressively more optimal samples. In Theta-Resonance, a neural network consumes a constant input tensor and produces a policy as a set of conditional probability density functions (PDFs) for sampling each design dimension. We specialize existing policy gradient algorithms in deep reinforcement learning (D-RL) in order to use evaluation feedback (in terms of cost, penalty or reward) to update our policy network with robust algorithmic stability and minimal design evaluations. We study multiple neural architectures (for our policy network) within the context of a simple SoC design space and propose a method of constructing synthetic space exploration problems to compare and improve design space exploration (DSE) algorithms. Although we only present categorical design spaces, we also outline how to use Theta-Resonance in order to explore continuous and mixed continuous-discrete design spaces.

Long horizon robot learning tasks with sparse rewards pose a significant challenge for current reinforcement learning algorithms. A key feature enabling humans to learn challenging control tasks is that they often receive expert intervention that enables them to understand the high-level structure of the task before mastering low-level control actions. We propose a framework for leveraging expert intervention to solve long-horizon reinforcement learning tasks. We consider \emph{option templates}, which are specifications encoding a potential option that can be trained using reinforcement learning. We formulate expert intervention as allowing the agent to execute option templates before learning an implementation. This enables them to use an option, before committing costly resources to learning it. We evaluate our approach on three challenging reinforcement learning problems, showing that it outperforms state-of-the-art approaches by two orders of magnitude. Videos of trained agents and our code can be found at: https://sites.google.com/view/stickymittens

Thanks to the advantages of flexible deployment and high mobility, unmanned aerial vehicles (UAVs) have been widely applied in the areas of disaster management, agricultural plant protection, environment monitoring and so on. With the development of UAV and sensor technologies, UAV assisted data collection for Internet of Things (IoT) has attracted increasing attentions. In this article, the scenarios and key technologies of UAV assisted data collection are comprehensively reviewed. First, we present the system model including the network model and mathematical model of UAV assisted data collection for IoT. Then, we review the key technologies including clustering of sensors, UAV data collection mode as well as joint path planning and resource allocation. Finally, the open problems are discussed from the perspectives of efficient multiple access as well as joint sensing and data collection. This article hopefully provides some guidelines and insights for researchers in the area of UAV assisted data collection for IoT.

Object packing by autonomous robots is an im-portant challenge in warehouses and logistics industry. Most conventional data-driven packing planning approaches focus on regular cuboid packing, which are usually heuristic and limit the practical use in realistic applications with everyday objects. In this paper, we propose a deep hierarchical reinforcement learning approach to simultaneously plan packing sequence and placement for irregular object packing. Specifically, the top manager network infers packing sequence from six principal view heightmaps of all objects, and then the bottom worker network receives heightmaps of the next object to predict the placement position and orientation. The two networks are trained hierarchically in a self-supervised Q-Learning framework, where the rewards are provided by the packing results based on the top height , object volume and placement stability in the box. The framework repeats sequence and placement planning iteratively until all objects have been packed into the box or no space is remained for unpacked items. We compare our approach with existing robotic packing methods for irregular objects in a physics simulator. Experiments show that our approach can pack more objects with less time cost than the state-of-the-art packing methods of irregular objects. We also implement our packing plan with a robotic manipulator to show the generalization ability in the real world.

Recent advances in the reinforcement learning (RL) literature have enabled roboticists to automatically train complex policies in simulated environments. However, due to the poor sample complexity of these methods, solving RL problems using real-world data remains a challenging problem. This paper introduces a novel cost-shaping method which aims to reduce the number of samples needed to learn a stabilizing controller. The method adds a term involving a Control Lyapunov Function (CLF) -- an `energy-like' function from the model-based control literature -- to typical cost formulations. Theoretical results demonstrate the new costs lead to stabilizing controllers when smaller discount factors are used, which is well-known to reduce sample complexity. Moreover, the addition of the CLF term `robustifies' the search for a stabilizing controller by ensuring that even highly sub-optimal polices will stabilize the system. We demonstrate our approach with two hardware examples where we learn stabilizing controllers for a cartpole and an A1 quadruped with only seconds and a few minutes of fine-tuning data, respectively. Furthermore, simulation benchmark studies show that obtaining stabilizing policies by optimizing our proposed costs requires orders of magnitude less data compared to standard cost designs.

Reinforcement learning has recently been used to approach well-known NP-hard combinatorial problems in graph theory. Among these problems, Hamiltonian cycle problems are exceptionally difficult to analyze, even when restricted to individual instances of structurally complex graphs. In this paper, we use Monte Carlo Tree Search (MCTS), the search algorithm behind many state-of-the-art reinforcement learning algorithms such as AlphaZero, to create autonomous agents that learn to play the game of Snake, a game centered on properties of Hamiltonian cycles on grid graphs. The game of Snake can be formulated as a single-player discounted Markov Decision Process (MDP) where the agent must behave optimally in a stochastic environment. Determining the optimal policy for Snake, defined as the policy that maximizes the probability of winning - or win rate - with higher priority and minimizes the expected number of time steps to win with lower priority, is conjectured to be NP-hard. Performance-wise, compared to prior work in the Snake game, our algorithm is the first to achieve a win rate over $0.5$ (a uniform random policy achieves a win rate $< 2.57 \times 10^{-15}$), demonstrating the versatility of AlphaZero in approaching NP-hard environments.