Value estimation is one key problem in Reinforcement Learning. Albeit many successes have been achieved by Deep Reinforcement Learning (DRL) in different fields, the underlying structure and learning dynamics of value function, especially with complex function approximation, are not fully understood. In this paper, we report that decreasing rank of $Q$-matrix widely exists during learning process across a series of continuous control tasks for different popular algorithms. We hypothesize that the low-rank phenomenon indicates the common learning dynamics of $Q$-matrix from stochastic high dimensional space to smooth low dimensional space. Moreover, we reveal a positive correlation between value matrix rank and value estimation uncertainty. Inspired by above evidence, we propose a novel Uncertainty-Aware Low-rank Q-matrix Estimation (UA-LQE) algorithm as a general framework to facilitate the learning of value function. Through quantifying the uncertainty of state-action value estimation, we selectively erase the entries of highly uncertain values in state-action value matrix and conduct low-rank matrix reconstruction for them to recover their values. Such a reconstruction exploits the underlying structure of value matrix to improve the value approximation, thus leading to a more efficient learning process of value function. In the experiments, we evaluate the efficacy of UA-LQE in several representative OpenAI MuJoCo continuous control tasks.

Operating in the real-world often requires agents to learn about a complex environment and apply this understanding to achieve a breadth of goals. This problem, known as goal-conditioned reinforcement learning (GCRL), becomes especially challenging for long-horizon goals. Current methods have tackled this problem by augmenting goal-conditioned policies with graph-based planning algorithms. However, they struggle to scale to large, high-dimensional state spaces and assume access to exploration mechanisms for efficiently collecting training data. In this work, we introduce Successor Feature Landmarks (SFL), a framework for exploring large, high-dimensional environments so as to obtain a policy that is proficient for any goal. SFL leverages the ability of successor features (SF) to capture transition dynamics, using it to drive exploration by estimating state-novelty and to enable high-level planning by abstracting the state-space as a non-parametric landmark-based graph. We further exploit SF to directly compute a goal-conditioned policy for inter-landmark traversal, which we use to execute plans to "frontier" landmarks at the edge of the explored state space. We show in our experiments on MiniGrid and ViZDoom that SFL enables efficient exploration of large, high-dimensional state spaces and outperforms state-of-the-art baselines on long-horizon GCRL tasks.

We propose the first near-optimal quantum algorithm for estimating in Euclidean norm the mean of a vector-valued random variable with finite mean and covariance. Our result aims at extending the theory of multivariate sub-Gaussian estimators to the quantum setting. Unlike classically, where any univariate estimator can be turned into a multivariate estimator with at most a logarithmic overhead in the dimension, no similar result can be proved in the quantum setting. Indeed, Heinrich ruled out the existence of a quantum advantage for the mean estimation problem when the sample complexity is smaller than the dimension. Our main result is to show that, outside this low-precision regime, there is a quantum estimator that outperforms any classical estimator. Our approach is substantially more involved than in the univariate setting, where most quantum estimators rely only on phase estimation. We exploit a variety of additional algorithmic techniques such as amplitude amplification, the Bernstein-Vazirani algorithm, and quantum singular value transformation. Our analysis also uses concentration inequalities for multivariate truncated statistics. We develop our quantum estimators in two different input models that showed up in the literature before. The first one provides coherent access to the binary representation of the random variable and it encompasses the classical setting. In the second model, the random variable is directly encoded into the phases of quantum registers. This model arises naturally in many quantum algorithms but it is often incomparable to having classical samples. We adapt our techniques to these two settings and we show that the second model is strictly weaker for solving the mean estimation problem. Finally, we describe several applications of our algorithms, notably in measuring the expectation values of commuting observables and in the field of machine learning.

This study is aimed at addressing the problem of fault tolerance of quadruped robots to actuator failure, which is critical for robots operating in remote or extreme environments. In particular, an adaptive curriculum reinforcement learning algorithm with dynamics randomization (ACDR) is established. The ACDR algorithm can adaptively train a quadruped robot in random actuator failure conditions and formulate a single robust policy for fault-tolerant robot control. It is noted that the hard2easy curriculum is more effective than the easy2hard curriculum for quadruped robot locomotion. The ACDR algorithm can be used to build a robot system that does not require additional modules for detecting actuator failures and switching policies. Experimental results show that the ACDR algorithm outperforms conventional algorithms in terms of the average reward and walking distance.

Real-world robotic tasks require complex reward functions. When we define the problem the robot needs to solve, we pretend that a designer specifies this complex reward exactly, and it is set in stone from then on. In practice, however, reward design is an iterative process: the designer chooses a reward, eventually encounters an "edge-case" environment where the reward incentivizes the wrong behavior, revises the reward, and repeats. What would it mean to rethink robotics problems to formally account for this iterative nature of reward design? We propose that the robot not take the specified reward for granted, but rather have uncertainty about it, and account for the future design iterations as future evidence. We contribute an Assisted Reward Design method that speeds up the design process by anticipating and influencing this future evidence: rather than letting the designer eventually encounter failure cases and revise the reward then, the method actively exposes the designer to such environments during the development phase. We test this method in a simplified autonomous driving task and find that it more quickly improves the car's behavior in held-out environments by proposing environments that are "edge cases" for the current reward.

In 2021 the Johns Hopkins University Applied Physics Laboratory held an internal challenge to develop artificially intelligent (AI) agents that could excel at the collaborative card game Hanabi. Agents were evaluated on their ability to play with human players whom the agents had never previously encountered. This study details the development of the agent that won the challenge by achieving a human-play average score of 16.5, outperforming the current state-of-the-art for human-bot Hanabi scores. The winning agent's development consisted of observing and accurately modeling the author's decision making in Hanabi, then training with a behavioral clone of the author. Notably, the agent discovered a human-complementary play style by first mimicking human decision making, then exploring variations to the human-like strategy that led to higher simulated human-bot scores. This work examines in detail the design and implementation of this human compatible Hanabi teammate, as well as the existence and implications of human-complementary strategies and how they may be explored for more successful applications of AI in human machine teams.

Increasing complexity of modern chips makes design validation more difficult. Existing approaches are not able anymore to cope with the complexity of tasks such as robust performance tuning in post-silicon validation. Therefore, we propose a novel approach based on learn-to-optimize and reinforcement learning in order to solve complex and mixed-type tuning tasks in a efficient and robust way.

Generative Flow Networks (GFlowNets) have been introduced as a method to sample a diverse set of candidates in an active learning context, with a training objective that makes them approximately sample in proportion to a given reward function. In this paper, we show a number of additional theoretical properties of GFlowNets. They can be used to estimate joint probability distributions and the corresponding marginal distributions where some variables are unspecified and, of particular interest, can represent distributions over composite objects like sets and graphs. GFlowNets amortize the work typically done by computationally expensive MCMC methods in a single but trained generative pass. They could also be used to estimate partition functions and free energies, conditional probabilities of supersets (supergraphs) given a subset (subgraph), as well as marginal distributions over all supersets (supergraphs) of a given set (graph). We introduce variations enabling the estimation of entropy and mutual information, sampling from a Pareto frontier, connections to reward-maximizing policies, and extensions to stochastic environments, continuous actions and modular energy functions.

Continuously learning new tasks using high-level ideas or knowledge is a key capability of humans. In this paper, we propose Lifelong reinforcement learning with Sequential linear temporal logic formulas and Reward Machines (LSRM), which enables an agent to leverage previously learned knowledge to fasten learning of logically specified tasks. For the sake of more flexible specification of tasks, we first introduce Sequential Linear Temporal Logic (SLTL), which is a supplement to the existing Linear Temporal Logic (LTL) formal language. We then utilize Reward Machines (RM) to exploit structural reward functions for tasks encoded with high-level events, and propose automatic extension of RM and efficient knowledge transfer over tasks for continuous learning in lifetime. Experimental results show that LSRM outperforms the methods that learn the target tasks from scratch by taking advantage of the task decomposition using SLTL and knowledge transfer over RM during the lifelong learning process.

Recently, Truncated Quantile Critics (TQC), using distributional representation of critics, was shown to provide state-of-the-art asymptotic training performance on all environments from the MuJoCo continuous control benchmark suite. Also recently, Randomized Ensemble Double Q-Learning (REDQ), using a high updateto-data ratio and target randomization, was shown to achieve high sample efficiency that is competitive with state-of-the-art model-based methods. In this paper, we propose a novel model-free algorithm, Aggressive Q-Learning with Ensembles (AQE), which improves the sample-efficiency performance of REDQ and the asymptotic performance of TQC, thereby providing overall state-of-the-art performance during all stages of training. Moreover, AQE is very simple, requiring neither distributional representation of critics nor target randomization. Off-policy Deep Reinforcement Learning algorithms aim to improve sample efficiency by reusing past experience. A number of off-policy Deep RL algorithms have been proposed for control tasks with continuous state and action spaces, including Deep Deterministic Policy Gradient (DDPG), Twin Delayed DDPG (TD3) and Soft Actor Critic (SAC) (Lillicrap et al., 2016; Fujimoto et al., 2018; Haarnoja et al., 2018a;b). TD3 introduced clipped double-Q learning, and was shown to be significantly more sample efficient than popular on-policy methods for a wide range of MuJoCo benchmarks. Soft Actor Critic (SAC) has similar off-policy structures with clipped double-Q learning, but it also employs maximum entropy reinforcement learning. SAC was shown to provide excellent sample efficiency and asymptotic performance in a wide-range of MuJoCo environments, including the high-dimensional Humanoid environment for which both DDPG and TD3 perform poorly.