Exploration is widely regarded as one of the most challenging aspects of reinforcement learning (RL), with many naive approaches succumbing to exponential sample complexity. To isolate the challenges of exploration, we propose a new "reward-free RL" framework. In the exploration phase, the agent first collects trajectories from an MDP $\mathcal{M}$ without a pre-specified reward function. After exploration, it is tasked with computing near-optimal policies under for $\mathcal{M}$ for a collection of given reward functions. This framework is particularly suitable when there are many reward functions of interest, or when the reward function is shaped by an external agent to elicit desired behavior. We give an efficient algorithm that conducts $\tilde{\mathcal{O}}(S^2A\mathrm{poly}(H)/\epsilon^2)$ episodes of exploration and returns $\epsilon$-suboptimal policies for an arbitrary number of reward functions. We achieve this by finding exploratory policies that visit each "significant" state with probability proportional to its maximum visitation probability under any possible policy. Moreover, our planning procedure can be instantiated by any black-box approximate planner, such as value iteration or natural policy gradient. We also give a nearly-matching $\Omega(S^2AH^2/\epsilon^2)$ lower bound, demonstrating the near-optimality of our algorithm in this setting.

In batch reinforcement learning (RL), one often constrains a learned policy to be close to the behavior (data-generating) policy, e.g., by constraining the learned action distribution to differ from the behavior policy by some maximum degree that is the same at each state. This can cause batch RL to be overly conservative, unable to exploit large policy changes at frequently-visited, high-confidence states without risking poor performance at sparsely-visited states. To remedy this, we propose residual policies, where the allowable deviation of the learned policy is state-action-dependent. We derive a new for RL method, BRPO, which learns both the policy and allowable deviation that jointly maximize a lower bound on policy performance. We show that BRPO achieves the state-of-the-art performance in a number of tasks.

There is broad interest in creating RL agents that can solve many (related) tasks and adapt to new tasks and environments after initial training. Model-based RL leverages learned surrogate models that describe dynamics and rewards of individual tasks, such that planning in a good surrogate can lead to good control of the true system. Rather than solving each task individually from scratch, hierarchical models can exploit the fact that tasks are often related by (unobserved) causal factors of variation in order to achieve efficient generalization, as in learning how the mass of an item affects the force required to lift it can generalize to previously unobserved masses. We propose Generalized Hidden Parameter MDPs (GHP-MDPs) that describe a family of MDPs where both dynamics and reward can change as a function of hidden parameters that vary across tasks. The GHP-MDP augments model-based RL with latent variables that capture these hidden parameters, facilitating transfer across tasks. We also explore a variant of the model that incorporates explicit latent structure mirroring the causal factors of variation across tasks (for instance: agent properties, environmental factors, and goals). We experimentally demonstrate state-of-the-art performance and sample-efficiency on a new challenging MuJoCo task using reward and dynamics latent spaces, while beating a previous state-of-the-art baseline with $>10\times$ less data. Using test-time inference of the latent variables, our approach generalizes in a single episode to novel combinations of dynamics and reward, and to novel rewards.

Transfer learning is an important new subfield of multiagent reinforcement learning that aims to help an agent learn about a problem by using knowledge that it has gained solving another problem, or by using knowledge that is communicated to it by an agent who already knows the problem. This is useful when one wishes to change the architecture or learning algorithm of an agent (so that the new knowledge need not be built "from scratch"), when new agents are frequently introduced to the environment with no knowledge, or when an agent must adapt to similar but different problems. Great progress has been made in the agent-to-agent case using the Teacher/Student framework proposed by (Torrey and Taylor 2013). However, that approach requires that learning from a teacher be treated differently from learning in every other reinforcement learning context. In this paper, I propose a method which allows the teacher/student framework to be applied in a way that fits directly and naturally into the more general reinforcement learning framework by integrating the teacher feedback into the reward signal received by the learning agent. I show that this approach can significantly improve the rate of learning for an agent playing a one-player stochastic game; I give examples of potential pitfalls of the approach; and I propose further areas of research building on this framework.

In reinforcement learning, we can learn a model of future observations and rewards, and use it to plan the agent's next actions. However, jointly modeling future observations can be computationally expensive or even intractable if the observations are high-dimensional (e.g. images). For this reason, previous works have considered partial models, which model only part of the observation. In this paper, we show that partial models can be causally incorrect: they are confounded by the observations they don't model, and can therefore lead to incorrect planning. To address this, we introduce a general family of partial models that are provably causally correct, yet remain fast because they do not need to fully model future observations.

The optimal policy of a reinforcement learning problem is often discontinuous and non-smooth. I.e., for two states with similar representations, their optimal policies can be significantly different. In this case, representing the entire policy with a function approximator (FA) with shared parameters for all states maybe not desirable, as the generalization ability of parameters sharing makes representing discontinuous, non-smooth policies difficult. A common way to solve this problem, known as Mixture-of-Experts, is to represent the policy as the weighted sum of multiple components, where different components perform well on different parts of the state space. Following this idea and inspired by a recent work called advantage-weighted information maximization, we propose to learn for each state weights of these components, so that they entail the information of the state itself and also the preferred action learned so far for the state. The action preference is characterized via the advantage function. In this case, the weight of each component would only be large for certain groups of states whose representations are similar and preferred action representations are also similar. Therefore each component is easy to be represented. We call a policy parameterized in this way an Advantage Weighted Mixture Policy (AWMP) and apply this idea to improve soft-actor-critic (SAC), one of the most competitive continuous control algorithm. Experimental results demonstrate that SAC with AWMP clearly outperforms SAC in four commonly used continuous control tasks and achieve stable performance across different random seeds.

Reinforcement learning has shown great promise in the training of robot behavior due to the sequential decision making characteristics. However, the required enormous amount of interactive and informative training data provides the major stumbling block for progress. In this study, we focus on accelerating reinforcement learning (RL) training and improving the performance of multi-goal reaching tasks. Specifically, we propose a precision-based continuous curriculum learning (PCCL) method in which the requirements are gradually adjusted during the training process, instead of fixing the parameter in a static schedule. To this end, we explore various continuous curriculum strategies for controlling a training process. This approach is tested using a Universal Robot 5e in both simulation and real-world multi-goal reach experiments. Experimental results support the hypothesis that a static training schedule is suboptimal, and using an appropriate decay function for curriculum learning provides superior results in a faster way.

Lane-change maneuvers are commonly executed by drivers to follow a certain routing plan, overtake a slower vehicle, adapt to a merging lane ahead, etc. However, improper lane change behaviors can be a major cause of traffic flow disruptions and even crashes. While many rule-based methods have been proposed to solve lane change problems for autonomous driving, they tend to exhibit limited performance due to the uncertainty and complexity of the driving environment. Machine learning-based methods offer an alternative approach, as Deep reinforcement learning (DRL) has shown promising success in many application domains including robotic manipulation, navigation, and playing video games. However, applying DRL for autonomous driving still faces many practical challenges in terms of slow learning rates, sample inefficiency, and non-stationary trajectories. In this study, we propose an automated lane change strategy using proximal policy optimization-based deep reinforcement learning, which shows great advantage in learning efficiency while maintaining stable performance. The trained agent is able to learn a smooth, safe, and efficient driving policy to determine lane-change decisions (i.e. when and how) even in dense traffic scenarios. The effectiveness of the proposed policy is validated using task success rate and collision rate, which demonstrates the lane change maneuvers can be efficiently learned and executed in a safe, smooth and efficient manner.

Tasks involving localization, memorization and planning in partially observable 3D environments are an ongoing challenge in Deep Reinforcement Learning. We present EgoMap, a spatially structured neural memory architecture. EgoMap augments a deep reinforcement learning agent's performance in 3D environments on challenging tasks with multi-step objectives. The EgoMap architecture incorporates several inductive biases including a differentiable inverse projection of CNN feature vectors onto a top-down spatially structured map. The map is updated with ego-motion measurements through a differentiable affine transform. We show this architecture outperforms both standard recurrent agents and state of the art agents with structured memory. We demonstrate that incorporating these inductive biases into an agent's architecture allows for stable training with reward alone, circumventing the expense of acquiring and labelling expert trajectories. A detailed ablation study demonstrates the impact of key aspects of the architecture and through extensive qualitative analysis, we show how the agent exploits its structured internal memory to achieve higher performance.