Exploration is critical for good results in deep reinforcement learning and has attracted much attention. However, existing multi-agent deep reinforcement learning algorithms still use mostly noise-based techniques. Very recently, exploration methods that consider cooperation among multiple agents have been developed. However, existing methods suffer from a common challenge: agents struggle to identify states that are worth exploring, and hardly coordinate exploration efforts toward those states. To address this shortcoming, in this paper, we propose cooperative multi-agent exploration (CMAE): agents share a common goal while exploring. The goal is selected from multiple projected state spaces via a normalized entropy-based technique. Then, agents are trained to reach this goal in a coordinated manner. We demonstrate that CMAE consistently outperforms baselines on various tasks, including a sparse-reward version of the multiple-particle environment (MPE) and the Starcraft multi-agent challenge (SMAC).

Abstract--Language instruction plays an essential role in the natural language grounded navigation tasks. However, navigators trained with limited human-annotated instructions may have difficulties in accurately capturing key information from the complicated instruction at different timesteps, leading to poor navigation performance. In this paper, we exploit to train a more robust navigator which is capable of dynamically extracting crucial factors from the long instruction, by using an adversarial attacking paradigm. Specifically, we propose a Dynamic Reinforced Instruction Attacker (DR-Attacker), which learns to mislead the navigator to move to the wrong target by destroying the most instructive information in instructions at different timesteps. By formulating the perturbation generation as a Markov Decision Process, DR-Attacker is optimized by the reinforcement learning algorithm to generate perturbed instructions sequentially during the navigation, according to a learnable attack score. Then, the perturbed instructions, which serve as hard samples, are used for improving the robustness of the navigator with an effective adversarial training strategy and an auxiliary self-supervised reasoning task. Experimental results on both Vision-and-Language Navigation (VLN) and Navigation from Dialog History (NDH) tasks show the superiority of our proposed method over state-of-the-art methods. Moreover, the visualization analysis shows the effectiveness of the proposed DR-Attacker, which can successfully attack crucial information in the instructions at different timesteps.

Deep Q Network (DQN) has several limitations when applied in planning a path in environment with a number of dilemmas according to our experiment. The reward function may be hard to model, and successful experience transitions are difficult to find in experience replay. In this context, this paper proposes an improved Double DQN (DDQN) to solve the problem by reference to A* and Rapidly-Exploring Random Tree (RRT). In order to achieve the rich experiments in experience replay, the initialization of robot in each training round is redefined based on RRT strategy. In addition, reward for the free positions is specially designed to accelerate the learning process according to the definition of position cost in A*. The simulation experimental results validate the efficiency of the improved DDQN, and robot could successfully learn the ability of obstacle avoidance and optimal path planning in which DQN or DDQN has no effect.

Two years ago, Open AI released Safety Gym, a suite of environments and tools for measuring progress towards reinforcement learning agents that respect safety constraints while training. Safety Gym has use cases across the reinforcement learning ecosystem. The open-source release is available on GitHub, where researchers and developers can get started with just a few lines of code. In this article, we will explore some of the alternative environments, tools and libraries for researchers to train machine learning models. AI Safety Gridworlds is a suite of reinforcement learning environments illustrating various safety properties of intelligent agents.

We present Bayesian Controller Fusion (BCF): a hybrid control strategy that combines the strengths of traditional hand-crafted controllers and model-free deep reinforcement learning (RL). BCF thrives in the robotics domain, where reliable but suboptimal control priors exist for many tasks, but RL from scratch remains unsafe and data-inefficient. By fusing uncertainty-aware distributional outputs from each system, BCF arbitrates control between them, exploiting their respective strengths. We study BCF on two real-world robotics tasks involving navigation in a vast and long-horizon environment, and a complex reaching task that involves manipulability maximisation. For both these domains, there exist simple handcrafted controllers that can solve the task at hand in a risk-averse manner but do not necessarily exhibit the optimal solution given limitations in analytical modelling, controller miscalibration and task variation. As exploration is naturally guided by the prior in the early stages of training, BCF accelerates learning, while substantially improving beyond the performance of the control prior, as the policy gains more experience. More importantly, given the risk-aversity of the control prior, BCF ensures safe exploration and deployment, where the control prior naturally dominates the action distribution in states unknown to the policy. We additionally show BCF's applicability to the zero-shot sim-to-real setting and its ability to deal with out-of-distribution states in the real-world. BCF is a promising approach for combining the complementary strengths of deep RL and traditional robotic control, surpassing what either can achieve independently. The code and supplementary video material are made publicly available at https://krishanrana.github.io/bcf.

Intrinsic rewards are commonly applied to improve exploration in reinforcement learning. However, these approaches suffer from instability caused by non-stationary reward shaping and strong dependency on hyperparameters. In this work, we propose Decoupled RL (DeRL) which trains separate policies for exploration and exploitation. DeRL can be applied with on-policy and off-policy RL algorithms. We evaluate DeRL algorithms in two sparse-reward environments with multiple types of intrinsic rewards. We show that DeRL is more robust to scaling and speed of decay of intrinsic rewards and converges to the same evaluation returns than intrinsically motivated baselines in fewer interactions.

In the stochastic linear contextual bandit setting there exist several minimax procedures for exploration with policies that are reactive to the data being acquired. In practice, there can be a significant engineering overhead to deploy these algorithms, especially when the dataset is collected in a distributed fashion or when a human in the loop is needed to implement a different policy. Exploring with a single non-reactive policy is beneficial in such cases. Assuming some batch contexts are available, we design a single stochastic policy to collect a good dataset from which a near-optimal policy can be extracted. We present a theoretical analysis as well as numerical experiments on both synthetic and real-world datasets.

Legged robots have been shown to be effective in navigating unstructured environments. Although there has been much success in learning locomotion policies for quadruped robots, there is little research on how to incorporate human knowledge to facilitate this learning process. In this paper, we demonstrate that human knowledge in the form of LTL formulas can be applied to quadruped locomotion learning within a Reward Machine (RM) framework. Experimental results in simulation show that our RM-based approach enables easily defining diverse locomotion styles, and efficiently learning locomotion policies of the defined styles.

Genomic selection (GS) is a technique that plant breeders use to select individuals to mate and produce new generations of species. Allocation of resources is a key factor in GS. At each selection cycle, breeders are facing the choice of budget allocation to make crosses and produce the next generation of breeding parents. Inspired by recent advances in reinforcement learning for AI problems, we develop a reinforcement learning-based algorithm to automatically learn to allocate limited resources across different generations of breeding. We mathematically formulate the problem in the framework of Markov Decision Process (MDP) by defining state and action spaces. To avoid the explosion of the state space, an integer linear program is proposed that quantifies the trade-off between resources and time. Finally, we propose a value function approximation method to estimate the action-value function and then develop a greedy policy improvement technique to find the optimal resources. We demonstrate the effectiveness of the proposed method in enhancing genetic gain using a case study with realistic data.

Can artificial agents learn to assist others in achieving their goals without knowing what those goals are? Generic reinforcement learning agents could be trained to behave altruistically towards others by rewarding them for altruistic behaviour, i.e., rewarding them for benefiting other agents in a given situation. Such an approach assumes that other agents' goals are known so that the altruistic agent can cooperate in achieving those goals. However, explicit knowledge of other agents' goals is often difficult to acquire. Even assuming such knowledge to be given, training of altruistic agents would require manually-tuned external rewards for each new environment. Thus, it is beneficial to develop agents that do not depend on external supervision and can learn altruistic behaviour in a task-agnostic manner. Assuming that other agents rationally pursue their goals, we hypothesize that giving them more choices will allow them to pursue those goals better. Some concrete examples include opening a door for others or safeguarding them to pursue their objectives without interference. We formalize this concept and propose an altruistic agent that learns to increase the choices another agent has by maximizing the number of states that the other agent can reach in its future. We evaluate our approach on three different multi-agent environments where another agent's success depends on the altruistic agent's behaviour. Finally, we show that our unsupervised agents can perform comparably to agents explicitly trained to work cooperatively. In some cases, our agents can even outperform the supervised ones.