Reinforcement learning is well-studied under discrete actions. Integer actions setting is popular in the industry yet still challenging due to its high dimensionality. To this end, we study reinforcement learning under integer actions by incorporating the Soft Actor-Critic (SAC) algorithm with an integer reparameterization. Our key observation for integer actions is that their discrete structure can be simplified using their comparability property. Hence, the proposed integer reparameterization does not need one-hot encoding and is of low dimensionality. Experiments show that the proposed SAC under integer actions is as good as the continuous action version on robot control tasks and outperforms Proximal Policy Optimization on power distribution systems control tasks.

There is a strong link between the general concept of intelligence and the ability to collect and use information. The theory of Bayes-adaptive exploration offers an attractive optimality framework for training machines to perform complex information gathering tasks. However, the computational complexity of the resulting optimal control problem has limited the diffusion of the theory to mainstream deep AI research. In this paper we exploit the inherent mathematical structure of Bayes-adaptive problems in order to dramatically simplify the problem by making the reward structure denser while simultaneously decoupling the learning of exploitation and exploration policies. The key to this simplification comes from the novel concept of cross-value (i.e. the value of being in an environment while acting optimally according to another), which we use to quantify the value of currently available information. This results in a new denser reward structure that "cashes in" all future rewards that can be predicted from the current information state. In a set of experiments we show that the approach makes it possible to learn challenging information gathering tasks without the use of shaping and heuristic bonuses in situations where the standard RL algorithms fail.

Since the earliest days of robotics, researchers have sought to build embodied agents to perform a variety of jobs, such as assistive tasks in factories [Oliff et al., 2020] or wildfire surveillance [Julian and Kochenderfer, 2019]. Following tremendous advancements in deep learning and convolutional neural networks in the past decade, researchers have been able to develop reinforcement learning (RL)-based embodied agents that interact with the real world on the basis of sensory observations. Software platforms such as OpenAI Gym [Brockman et al., 2016], Unity ML-Agents Toolkit [Juliani et al., 2018], and AI Habitat [Savva et al., 2019] have emerged to address the community's need for training and evaluating RL-based embodied agents end-to-end. Our research group was particularly intrigued by the AI Habitat platform, which offers a high-performance, photorealistic simulator, access to a sizeable library of visually-rich scanned 3D environments, and a modular software design. However, even though these platforms allow roboticists to reuse existing RL algorithms and train agents in simulators with ease, there is a critical step to using them for embodied agents which is only partially addressed: Connecting the trained agent with a real robot. Ideally, after training an RL agent in simulation one would like to take advantage of the extensive set of tools and knowledge from the robotics community to make it easy to embody that agent. One particularly popular tool from the robotics community is ROS, a robotics-focused middleware platform with extensive support for classical robotic mapping, planning and control algorithms ([mov, dwa]) as well as drivers for a wide variety of compute, sensing and actuation hardware. But ROS support for directly training an RL agent is limited, and Gazebo-- the standard simulation environment used for ROS systems-- cannot match the level of photorealism or simulation speed of tools specifically designed to train large-scale RL agents [Liang et al., 2019].

This work studies object goal navigation task, which involves navigating to the closest object related to the given semantic category in unseen environments. Recent works have shown significant achievements both in the end-to-end Reinforcement Learning approach and modular systems, but need a big step forward to be robust and optimal. We propose a hierarchical method that incorporates standard task formulation and additional area knowledge as landmarks, with a way to extract these landmarks. In a hierarchy, a low level consists of separately trained algorithms to the most intuitive skills, and a high level decides which skill is needed at this moment. With all proposed solutions, we achieve a 0.75 success rate in a realistic Habitat simulator. After a small stage of additional model training in a reconstructed virtual area at a simulator, we successfully confirmed our results in a real-world case.

While the rapid progress of deep learning fuels end-to-end reinforcement learning (RL), direct application, especially in high-dimensional space like robotic scenarios still suffers from high sample efficiency. Therefore State Representation Learning (SRL) is proposed to specifically learn to encode task-relevant features from complex sensory data into low-dimensional states. However, the pervasive implementation of SRL is usually conducted by a decoupling strategy in which the observation-state mapping is learned separately, which is prone to over-fit. To handle such problem, we present a new algorithm called Policy Optimization via Abstract Representation which integrates SRL into the original RL scale. Firstly, We engage RL loss to assist in updating SRL model so that the states can evolve to meet the demand of reinforcement learning and maintain a good physical interpretation. Secondly, we introduce a dynamic parameter adjustment mechanism so that both models can efficiently adapt to each other. Thirdly, we introduce a new prior called domain resemblance to leverage expert demonstration to train the SRL model. Finally, we provide a real-time access by state graph to monitor the course of learning. Results show that our algorithm outperforms the PPO baselines and decoupling strategies in terms of sample efficiency and final rewards. Thus our model can efficiently deal with tasks in high dimensions and facilitate training real-life robots directly from scratch.

The challenge of mapping indoor environments is addressed. Typical heuristic algorithms for solving the motion planning problem are frontier-based methods, that are especially effective when the environment is completely unknown. However, in cases where prior statistical data on the environment's architectonic features is available, such algorithms can be far from optimal. Furthermore, their calculation time may increase substantially as more areas are exposed. In this paper we propose two means by which to overcome these shortcomings. One is the use of deep reinforcement learning to train the motion planner. The second is the inclusion of a pre-trained generative deep neural network, acting as a map predictor. Each one helps to improve the decision making through use of the learned structural statistics of the environment, and both, being realized as neural networks, ensure a constant calculation time. We show that combining the two methods can shorten the mapping time, compared to frontier-based motion planning, by up to 75%.

Autonomous driving in urban crowds at unregulated intersections is challenging, where dynamic occlusions and uncertain behaviors of other vehicles should be carefully considered. Traditional methods are heuristic and based on hand-engineered rules and parameters, but scale poorly in new situations. Therefore, they require high labor cost to design and maintain rules in all foreseeable scenarios. Recently, deep reinforcement learning (DRL) has shown promising results in urban driving scenarios. However, DRL is known to be sample inefficient, and most previous works assume perfect observations such as ground-truth locations and motions of vehicles without considering noises and occlusions, which might be a too strong assumption for policy deployment. In this work, we use DRL to train lidar-based end-to-end driving policies that naturally consider imperfect partial observations. We further use unsupervised contrastive representation learning as an auxiliary task to improve the sample efficiency. The comparative evaluation results reveal that our method achieves higher success rates than the state-of-the-art (SOTA) lidar-based end-to-end driving network, better trades off safety and efficiency than the carefully tuned rule-based method, and generalizes better to new scenarios than the baselines. Demo videos are available at https://caipeide.github.io/carl-lead/.

Effective robot learning often requires online human feedback and interventions that can cost significant human time, giving rise to the central challenge in interactive imitation learning: is it possible to control the timing and length of interventions to both facilitate learning and limit burden on the human supervisor? This paper presents ThriftyDAgger, an algorithm for actively querying a human supervisor given a desired budget of human interventions. ThriftyDAgger uses a learned switching policy to solicit interventions only at states that are sufficiently (1) novel, where the robot policy has no reference behavior to imitate, or (2) risky, where the robot has low confidence in task completion. To detect the latter, we introduce a novel metric for estimating risk under the current robot policy. Experiments in simulation and on a physical cable routing experiment suggest that ThriftyDAgger's intervention criteria balances task performance and supervisor burden more effectively than prior algorithms. ThriftyDAgger can also be applied at execution time, where it achieves a 100% success rate on both the simulation and physical tasks. A user study (N=10) in which users control a three-robot fleet while also performing a concentration task suggests that ThriftyDAgger increases human and robot performance by 58% and 80% respectively compared to the next best algorithm while reducing supervisor burden.

Interest in applying Artificial Intelligence (AI) techniques to compiler optimizations is increasing rapidly, but compiler research has a high entry barrier. Unlike in other domains, compiler and AI researchers do not have access to the datasets and frameworks that enable fast iteration and development of ideas, and getting started requires a significant engineering investment. What is needed is an easy, reusable experimental infrastructure for real world compiler optimization tasks that can serve as a common benchmark for comparing techniques, and as a platform to accelerate progress in the field. We introduce CompilerGym, a set of environments for real world compiler optimization tasks, and a toolkit for exposing new optimization tasks to compiler researchers. CompilerGym enables anyone to experiment on production compiler optimization problems through an easy-to-use package, regardless of their experience with compilers. We build upon the popular OpenAI Gym interface enabling researchers to interact with compilers using Python and a familiar API. We describe the CompilerGym architecture and implementation, characterize the optimization spaces and computational efficiencies of three included compiler environments, and provide extensive empirical evaluations. Compared to prior works, CompilerGym offers larger datasets and optimization spaces, is 27x more computationally efficient, is fault-tolerant, and capable of detecting reproducibility bugs in the underlying compilers. In making it easy for anyone to experiment with compilers - irrespective of their background - we aim to accelerate progress in the AI and compiler research domains.

The growing number of applications of Reinforcement Learning (RL) in real-world domains has led to the development of privacy-preserving techniques due to the inherently sensitive nature of data. Most existing works focus on differential privacy, in which information is revealed in the clear to an agent whose learned model should be robust against information leakage to malicious third parties. Motivated by use cases in which only encrypted data might be shared, such as information from sensitive sites, in this work we consider scenarios in which the inputs themselves are sensitive and cannot be revealed. We develop a simple extension to the MDP framework which provides for the encryption of states. We present a preliminary, experimental study of how a DQN agent trained on encrypted states performs in environments with discrete and continuous state spaces. Our results highlight that the agent is still capable of learning in small state spaces even in presence of non-deterministic encryption, but performance collapses in more complex environments.