The actor-critic (AC) reinforcement learning algorithms have been the powerhouse behind many challenging applications. Nevertheless, its convergence is fragile in general. To study its instability, existing works mostly consider the uncommon double-loop variant or basic models with finite state and action space. We investigate the more practical single-sample two-timescale AC for solving the canonical linear quadratic regulator (LQR) problem, where the actor and the critic update only once with a single sample in each iteration on an unbounded continuous state and action space. Existing analysis cannot conclude the convergence for such a challenging case. We develop a new analysis framework that allows establishing the global convergence to an $\epsilon$-optimal solution with at most an $\mathcal{O}(\epsilon^{-2.5})$ sample complexity. To our knowledge, this is the first finite-time convergence analysis for the single sample two-timescale AC for solving LQR with global optimality. The sample complexity improves those of other variants by orders, which sheds light on the practical wisdom of single sample algorithms. We also further validate our theoretical findings via comprehensive simulation comparisons.

Learning-based methods in robotics hold the promise of generalization, but what can be done if a learned policy does not generalize to a new situation? In principle, if an agent can at least evaluate its own success (i.e., with a reward classifier that generalizes well even when the policy does not), it could actively practice the task and finetune the policy in this situation. We study this problem in the setting of industrial insertion tasks, such as inserting connectors in sockets and setting screws. Existing algorithms rely on precise localization of the connector or socket and carefully managed physical setups, such as assembly lines, to succeed at the task. But in unstructured environments such as homes or even some industrial settings, robots cannot rely on precise localization and may be tasked with previously unseen connectors. Offline reinforcement learning on a variety of connector insertion tasks is a potential solution, but what if the robot is tasked with inserting previously unseen connector? In such a scenario, we will still need methods that can robustly solve such tasks with online practice. One of the main observations we make in this work is that, with a suitable representation learning and domain generalization approach, it can be significantly easier for the reward function to generalize to a new but structurally similar task (e.g., inserting a new type of connector) than for the policy. This means that a learned reward function can be used to facilitate the finetuning of the robot's policy in situations where the policy fails to generalize in zero shot, but the reward function generalizes successfully. We show that such an approach can be instantiated in the real world, pretrained on 50 different connectors, and successfully finetuned to new connectors via the learned reward function. Videos can be viewed at https://sites.google.com/view/learningonthejob

We develop novel methodology for active feature acquisition (AFA), the study of how to sequentially acquire a dynamic (on a per instance basis) subset of features that minimizes acquisition costs whilst still yielding accurate predictions. The AFA framework can be useful in a myriad of domains, including health care applications where the cost of acquiring additional features for a patient (in terms of time, money, risk, etc.) can be weighed against the expected improvement to diagnostic performance. Previous approaches for AFA have employed either: deep learning RL techniques, which have difficulty training policies in the AFA MDP due to sparse rewards and a complicated action space; deep learning surrogate generative models, which require modeling complicated multidimensional conditional distributions; or greedy policies, which fail to account for how joint feature acquisitions can be informative together for better predictions. In this work we show that we can bypass many of these challenges with a novel, nonparametric oracle based approach, which we coin the acquisition conditioned oracle (ACO). Extensive experiments show the superiority of the ACO to state-of-the-art AFA methods when acquiring features for both predictions and general decision-making.

Offline reinforcement learning (RL) struggles in environments with rich and noisy inputs, where the agent only has access to a fixed dataset without environment interactions. Past works have proposed common workarounds based on the pre-training of state representations, followed by policy training. In this work, we introduce a simple, yet effective approach for learning state representations. Our method, Behavior Prior Representation (BPR), learns state representations with an easy-to-integrate objective based on behavior cloning of the dataset: we first learn a state representation by mimicking actions from the dataset, and then train a policy on top of the fixed representation, using any off-the-shelf Offline RL algorithm. Theoretically, we prove that BPR carries out performance guarantees when integrated into algorithms that have either policy improvement guarantees (conservative algorithms) or produce lower bounds of the policy values (pessimistic algorithms). Empirically, we show that BPR combined with existing state-of-the-art Offline RL algorithms leads to significant improvements across several offline control benchmarks. The code is available at \url{https://github.com/bit1029public/offline_bpr}.

Reinforcement Learning (RL) agents are often unable to generalise well to environment variations in the state space that were not observed during training. This issue is especially problematic for image-based RL, where a change in just one variable, such as the background colour, can change many pixels in the image. The changed pixels can lead to drastic changes in the agent's latent representation of the image, causing the learned policy to fail. To learn more robust representations, we introduce TEmporal Disentanglement (TED), a self-supervised auxiliary task that leads to disentangled image representations exploiting the sequential nature of RL observations. We find empirically that RL algorithms utilising TED as an auxiliary task adapt more quickly to changes in environment variables with continued training compared to state-of-the-art representation learning methods. Since TED enforces a disentangled structure of the representation, our experiments also show that policies trained with TED generalise better to unseen values of variables irrelevant to the task (e.g. background colour) as well as unseen values of variables that affect the optimal policy (e.g. goal positions).

Lee et al. (2020) utilize relative-attribute information in robot skill learning, but their GAN-based formulation is restricted to static visual attributes and is not applicable to temporally-extended concepts. This paper adopts a similar setup to works that learn diverse skills or motion styles from largescale offline behavior datasets or demonstrations (Lee & Popović, 2010; Wang et al., 2017; Zhou & Dragan, 2018; Peng et al., 2018b; Luo et al., 2020; Chebotar et al., 2021; Peng et al., 2021). These works emphasize on modeling a variety of reusable motor skills by learning a low-level controller conditioned on skill latent codes. Since the latent codes are inscrutable to humans, for each new task, the user must specify the desirable agent behavior by constructing an engineered symbolic reward and use it to train a separate high-level policy that controls the low-level controller. Our methods are complemented by existing diverse-skill learning methods because skill priors (i.e., pre-trained low-level controllers) allow us to optimize the behavioral reward more efficiently. More recently, there have been works in diffusion-based text-to-motion animation generation (Tevet et al., 2022; Guo et al., 2022). They are similar to this work in the sense that we both allow humans to control the agent behavior through explicit concepts. However, they do not support fine-grained control over the strength of individual behavioral attributes, and their works are not applicable to physics-based character control.

Traffic signal control is safety-critical for our daily life. Roughly one-quarter of road accidents in the U.S. happen at intersections due to problematic signal timing, urging the development of safety-oriented intersection control. However, existing studies on adaptive traffic signal control using reinforcement learning technologies have focused mainly on minimizing traffic delay but neglecting the potential exposure to unsafe conditions. We, for the first time, incorporate road safety standards as enforcement to ensure the safety of existing reinforcement learning methods, aiming toward operating intersections with zero collisions. We have proposed a safety-enhanced residual reinforcement learning method (SafeLight) and employed multiple optimization techniques, such as multi-objective loss function and reward shaping for better knowledge integration. Extensive experiments are conducted using both synthetic and real-world benchmark datasets. Results show that our method can significantly reduce collisions while increasing traffic mobility.

In offline reinforcement learning, weighted regression is a common method to ensure the learned policy stays close to the behavior policy and to prevent selecting out-of-sample actions. In this work, we show that due to the limited distributional expressivity of policy models, previous methods might still select unseen actions during training, which deviates from their initial motivation. To address this problem, we adopt a generative approach by decoupling the learned policy into two parts: an expressive generative behavior model and an action evaluation model. The key insight is that such decoupling avoids learning an explicitly parameterized policy model with a closed-form expression. Directly learning the behavior policy allows us to leverage existing advances in generative modeling, such as diffusion-based methods, to model diverse behaviors. As for action evaluation, we combine our method with an in-sample planning technique to further avoid selecting out-of-sample actions and increase computational efficiency. Experimental results on D4RL datasets show that our proposed method achieves competitive or superior performance compared with state-of-the-art offline RL methods, especially in complex tasks such as AntMaze. We also empirically demonstrate that our method can successfully learn from a heterogeneous dataset containing multiple distinctive but similarly successful strategies, whereas previous unimodal policies fail.

Consider a decision-maker that can pick one out of $K$ actions to control an unknown system, for $T$ turns. The actions are interpreted as different configurations or policies. Holding the same action fixed, the system asymptotically converges to a unique equilibrium, as a function of this action. The dynamics of the system are unknown to the decision-maker, which can only observe a noisy reward at the end of every turn. The decision-maker wants to maximize its accumulated reward over the $T$ turns. Learning what equilibria are better results in higher rewards, but waiting for the system to converge to equilibrium costs valuable time. Existing bandit algorithms, either stochastic or adversarial, achieve linear (trivial) regret for this problem. We present a novel algorithm, termed Upper Equilibrium Concentration Bound (UECB), that knows to switch an action quickly if it is not worth it to wait until the equilibrium is reached. This is enabled by employing convergence bounds to determine how far the system is from equilibrium. We prove that UECB achieves a regret of $\mathcal{O}(\log(T)+\tau_c\log(\tau_c)+\tau_c\log\log(T))$ for this equilibrium bandit problem where $\tau_c$ is the worst case approximate convergence time to equilibrium. We then show that both epidemic control and game control are special cases of equilibrium bandits, where $\tau_c\log \tau_c$ typically dominates the regret. We then test UECB numerically for both of these applications.

Off-policy learning ability is an important feature of reinforcement learning (RL) for practical applications. However, even one of the most elementary RL algorithms, temporal-difference (TD) learning, is known to suffer form divergence issue when the off-policy scheme is used together with linear function approximation. To overcome the divergent behavior, several off-policy TD-learning algorithms, including gradient-TD learning (GTD), and TD-learning with correction (TDC), have been developed until now. In this work, we provide a unified view of such algorithms from a purely control-theoretic perspective, and propose a new convergent algorithm. Our method relies on the backstepping technique, which is widely used in nonlinear control theory. Finally, convergence of the proposed algorithm is experimentally verified in environments where the standard TD-learning is known to be unstable.