Achieving human-level performance on real world tasks is a north star for the robotics community. We present the first learned robot agent that reaches amateur humanlevel performance in competitive table tennis. Table tennis is a physically demanding sport that takes humans years to master. We contribute (1) a hierarchical and modular policy architecture consisting of (i) low level controllers with their skill descriptors that model their capabilities and (ii) a high level controller that chooses the low level skills, (2) techniques for enabling zero-shot sim-to-real and curriculum building, including an iterative approach (train in sim, deploy in real), and (3) real time adaptation to unseen opponents. Policy performance was assessed through 29 robot vs. human matches of which the robot won 45 % (13/29). All humans were unseen players and their skill level varied from beginner to tournament level. Whilst the robot lost all matches vs. the most advanced players it won 100 % matches vs. beginners and 55 % matches vs. intermediate players, demonstrating solidly amateur humanlevel performance. Videos of the matches can be viewed here1.See sites https://google.com/view/competitive-robot-table-tennis.

We study the non-stationary stochastic multi-armed bandit problem, where the reward statistics of each arm may change several times during the course of learning. The performance of a learning algorithm is evaluated in terms of its dynamic regret, which is defined as the difference between the expected cumulative reward of an agent choosing the optimal arm in every time step and the cumulative reward of the learning algorithm. One way to measure the hardness of such environments is to consider how many times the identity of the optimal arm can change. We propose a method that achieves, in $K$-armed bandit problems, a near-optimal $\widetilde O(\sqrt{K N(S+1)})$ dynamic regret, where $N$ is the time horizon of the problem and $S$ is the number of times the identity of the optimal arm changes, without prior knowledge of $S$. Previous works for this problem obtain regret bounds that scale with the number of changes (or the amount of change) in the reward functions, which can be much larger, or assume prior knowledge of $S$ to achieve similar bounds.

Learning with simulators is ubiquitous in mod-ern reinforcement learning (RL). The simulatorcan either correspond to a simplified version ofthe real environment (such as a physics simulation of a robot arm) or to the environment itself (such as in games like Atari and Go). Among algorithms that are provably sample-efficient in this setting, most make the unrealistic assumption that all possible environment states are known before learning begins, or perform global optimistic planning which is computationally inefficient. In this work, we focus on simulation-based RL under a more realistic local access protocol, where the state space is unknown and the simulator can only be queried at states that have previously been observed (initial states and those returned by previous queries). We propose an algorithm named CONFIDENT-LSVI based on the template of least-square value iteration. CONFIDENT-LSVI incrementally builds a coreset of important states and uses the simulator to revisit them. Assuming that the linear function class has low approximation error under the Bell-man optimality operator (a.k.a. low inherent Bell-man error), we bound the algorithm performance in terms of this error, and show that it is query-and computationally-efficient.

We study query and computationally efficient planning algorithms for discounted Markov decision processes (MDPs) with linear function approximation and a simulator. The agent is assumed to have local access to the simulator, meaning that the simulator can be queried only at states that have been encountered in previous steps. We propose two new algorithms for this setting, which we call confident Monte Carlo least-squares policy iteration (Confident MC-LSPI), and confident Monte Carlo Politex (Confident MC-Politex), respectively. The main novelty in our algorithms is that they gradually build a set of state-action pairs (“core set”) with which it can control the extrapolation errors. We show that our algorithms have polynomial query and computational cost in the dimension of the features, the effective planning horizon and the targeted sub-optimality, while the cost remains independent of the size of the state space. An interesting technical contribution of our work is the introduction of a novel proof technique that makes use of a virtual policy iteration algorithm. We use this method to leverage existing results on approximate policy iteration with $\ell_\infty$-bounded error to show that our algorithm can learn the optimal policy for the given initial state even only with local access to the simulator. We believe that this technique can be extended to broader settings beyond this work.

Model-free reinforcement learning algorithms combined with value function approximation have recently achieved impressive performance in a variety of application domains. However, the theoretical understanding of such algorithms is limited, and existing results are largely focused on episodic or discounted Markov decision processes (MDPs). In this work, we present adaptive approximate policy iteration (AAPI), a learning scheme which enjoys a O(T^{2/3}) regret bound for undiscounted, continuing learning in uniformly ergodic MDPs. This is an improvement over the best existing bound of O(T^{3/4}) for the average-reward case with function approximation. Our algorithm and analysis rely on online learning techniques, where value functions are treated as losses. The main technical novelty is the use of a data-dependent adaptive learning rate coupled with a so-called optimistic prediction of upcoming losses. In addition to theoretical guarantees, we demonstrate the advantages of our approach empirically on several environments.

In this work, we study algorithms for learning in infinite-horizon undiscounted Markov decision processes (MDPs) with function approximation. We first show that the regret analysis of the Politex algorithm (a version of regularized policy iteration) can be sharpened from $O(T^{3/4})$ to $O(\sqrt{T})$ under nearly identical assumptions, and instantiate the bound with linear function approximation. Our result provides the first high-probability $O(\sqrt{T})$ regret bound for a computationally efficient algorithm in this setting. The exact implementation of Politex with neural network function approximation is inefficient in terms of memory and computation. Since our analysis suggests that we need to approximate the average of the action-value functions of past policies well, we propose a simple efficient implementation where we train a single Q-function on a replay buffer with past data. We show that this often leads to superior performance over other implementation choices, especially in terms of wall-clock time. Our work also provides a novel theoretical justification for using experience replay within policy iteration algorithms.

We propose a model-free algorithm for learning efficient policies capable of returning table tennis balls by controlling robot joints at a rate of 100Hz. We demonstrate that evolutionary search (ES) methods acting on CNN-based policy architectures for non-visual inputs and convolving across time learn compact controllers leading to smooth motions. Furthermore, we show that with appropriately tuned curriculum learning on the task and rewards, policies are capable of developing multi-modal styles, specifically forehand and backhand stroke, whilst achieving 80% return rate on a wide range of ball throws. We observe that multi-modality does not require any architectural priors, such as multi-head architectures or hierarchical policies.

This work focuses on off-policy evaluation (OPE) with function approximation in infinite-horizon undiscounted Markov decision processes (MDPs). For MDPs that are ergodic and linear (i.e. where rewards and dynamics are linear in some known features), we provide the first finite-sample OPE error bound, extending the existing results beyond the episodic and discounted cases. In a more general setting, when the feature dynamics are approximately linear and for arbitrary rewards, we propose a new approach for estimating stationary distributions with function approximation. We formulate this problem as finding the maximum-entropy distribution subject to matching feature expectations under empirical dynamics. We show that this results in an exponential-family distribution whose sufficient statistics are the features, paralleling maximum-entropy approaches in supervised learning. We demonstrate the effectiveness of the proposed OPE approaches in multiple environments.

Model-free approaches for reinforcement learning (RL) and continuous control find policies based only on past states and rewards, without fitting a model of the system dynamics. They are appealing as they are general purpose and easy to implement; however, they also come with fewer theoretical guarantees than model-based RL. In this work, we present a new model-free algorithm for controlling linear quadratic (LQ) systems, and show that its regret scales as $O(T^{\xi+2/3})$ for any small $\xi>0$ if time horizon satisfies $T>C^{1/\xi}$ for a constant $C$. The algorithm is based on a reduction of control of Markov decision processes to an expert prediction problem. In practice, it corresponds to a variant of policy iteration with forced exploration, where the policy in each phase is greedy with respect to the average of all previous value functions. This is the first model-free algorithm for adaptive control of LQ systems that provably achieves sublinear regret and has a polynomial computation cost. Empirically, our algorithm dramatically outperforms standard policy iteration, but performs worse than a model-based approach.

We study the problem of controlling linear time-invariant systems with known noisy dynamics and adversarially chosen quadratic losses. We present the first efficient online learning algorithms in this setting that guarantee $O(\sqrt{T})$ regret under mild assumptions, where $T$ is the time horizon. Our algorithms rely on a novel SDP relaxation for the steady-state distribution of the system. Crucially, and in contrast to previously proposed relaxations, the feasible solutions of our SDP all correspond to “strongly stable” policies that mix exponentially fast to a steady state.

Despite impressive recent advances in reinforcement learning (RL), its deployment in real-world physical systems is often complicated by unexpected events, limited data, and the potential for expensive failures. In this paper, we describe an application of RL “in the wild” to the task of regulating temperatures and airflow inside a large-scale data center (DC). Adopting a data-driven, model-based approach, we demonstrate that an RL agent with little prior knowledge is able to effectively and safely regulate conditions on a server floor after just a few hours of exploration, while improving operational efficiency relative to existing PID controllers.

Amir Globerson, Nevena Lazic, Soumen Chakrabarti, Amarnag Subramanya, Michael Ringgaard, Fernando Pereira. Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2016.

We present Plato, a probabilistic model for entity resolution that includes a novel approach for handling noisy or uninformative features, and supplements labeled training data derived from Wikipedia with a very large unlabeled text corpus. Training and inference in the proposed model can easily be distributed across many servers, allowing it to scale to over 107 entities. We evaluate Plato on three standard datasets for entity resolution. Our approach achieves the best results to-date on TAC KBP 2011 and is highly competitive on both the CoNLL 2003 and TAC KBP 2012 datasets.

Learning the structure of discrete Bayesian networks has been the subject of extensive research in machine learning, with most Bayesian approaches focusing on fully observed networks. One of few the methods that can handle networks with latent variables is the "structural EM algorithm" which interleaves greedy structure search with the estimation of latent variables and parameters, maintaining a single best network at each step. We introduce Structural Expectation Propagation (SEP), an extension of EP which can infer the structure of Bayesian networks having latent variables and missing data. SEP performs variational inference in a joint model of structure, latent variables, and parameters, offering two advantages: (i) it accounts for uncertainty in structure and parameter values when making local distribution updates (ii) it returns a variational distribution over network structures rather than a single network. We demonstrate the performance of SEP both on synthetic problems and on real-world clinical data.

Subspace segmentation is the task of segmenting data lying on multiple linear subspaces. Its applications in computer vision include motion segmentation in video, structure-from-motion, and image clustering. In this work, we describe a novel approach for subspace segmentation that uses probabilistic inference via a message-passing algorithm. We cast the subspace segmentation problem as that of choosing the best subset of linear subspaces from a set of candidate subspaces constructed from the data. Under this formulation, subspace segmentation corresponds to facility location, a well studied operational research problem. Approximate solutions to this NP-hard optimization problem can be found by performing maximum-a-posteriori (MAP) inference in a probabilistic graphical model. We describe the graphical model and a message-passing inference algorithm. We demonstrate the performance of Facility Location for Subspace Segmentation, or FLoSS, on synthetic data as well as on 3D multi-body video motion segmentation from point correspondences.

This work proposes a novel spoken term detection technique, where the query is in audio format. Detection and retrieval are performed by matching the spectrograms of the spoken document and query as visual images, using ideas from computer vision. Local descriptors are computed on a dense grid over each spectrogram, and the query term is detected using deformable template matching of grids. Detection experiments are performed on an hour-long newscast recording, involving 10 query terms of length 2-3 words. When the query term comes from the document, nearly all other instances of the term in the document are detected; performance degrades when the query is recorded by the user.

The impact of image feature locations in the bag-of-words model for object classification is examined. It is demonstrated that a simple variance-based method works well and offers advantages over several other methods. In essence, the feature locations are selected intelligently, decreasing the redundancy and cost sometimes associated with feature extraction on dense grids. Classification results on two databases are presented, using a support vector machine classifier.

This work proposes an audio data hiding system that hides information into signals not known beforehand. The system can hide data into live music, or ambient sounds in general, and can be used to communicate information acoustically from one device to another. An important benefit of such a system is backward-compatibility, as the transmitter is a speaker, and the receiver is a microphone, both of which are already present in numerous devices and environments. The highest data rate achieved was 213 bits/s, while keeping the error rate under 10%. The developed technique was applied in a simple navigation system, where acoustic data embedded into background music indicates the location of the receiver