The performance of imitation learning is typically upper-bounded by the performance of the demonstrator. Recent empirical results show that imitation learning via ranked demonstrations allows for better-than-demonstrator performance; however, ranked demonstrations may be difficult to obtain, and little is known theoretically about when such methods can be expected to outperform the demonstrator. To address these issues, we first contribute a sufficient condition for when better-than-demonstrator performance is possible and discuss why ranked demonstrations can contribute to better-than-demonstrator performance. Building on this theory, we then introduce Disturbance-based Reward Extrapolation (D-REX), a ranking-based imitation learning method that injects noise into a policy learned through behavioral cloning to automatically generate ranked demonstrations. By generating rankings automatically, ranking-based imitation learning can be applied in traditional imitation learning settings where only unlabeled demonstrations are available. We empirically validate our approach on standard MuJoCo and Atari benchmarks and show that D-REX can utilize automatic rankings to significantly surpass the performance of the demonstrator and outperform standard imitation learning approaches. D-REX is the first imitation learning approach to achieve significant extrapolation beyond the demonstrator's performance without additional side-information or supervision, such as rewards or human preferences.

Imitation learning, followed by reinforcement learning algorithms, is a promising paradigm to solve complex control tasks sample-efficiently. However, learning from demonstrations often suffers from the covariate shift problem, which results in cascading errors of the learned policy. We introduce a notion of conservatively-extrapolated value functions, which provably lead to policies with self-correction. We design an algorithm Value Iteration with Negative Sampling (VINS) that practically learns such value functions with conservative extrapolation. We show that VINS can correct mistakes of the behavioral cloning policy on simulated robotics benchmark tasks. We also propose the algorithm of using VINS to initialize a reinforcement learning algorithm, which is shown to outperform significantly prior works in sample efficiency.

This paper addresses two challenges facing sampling-based kinodynamic motion planning: a way to identify good candidate states for local transitions and the subsequent computationally intractable steering between these candidate states. Through the combination of sampling-based planning, a Rapidly Exploring Randomized Tree (RRT) and an efficient kinodynamic motion planner through machine learning, we propose an efficient solution to long-range planning for kinodynamic motion planning. First, we use deep reinforcement learning to learn an obstacle-avoiding policy that maps a robot's sensor observations to actions, which is used as a local planner during planning and as a controller during execution. Second, we train a reachability estimator in a supervised manner, which predicts the RL policy's time to reach a state in the presence of obstacles. Lastly, we introduce RL-RRT that uses the RL policy as a local planner, and the reachability estimator as the distance function to bias tree-growth towards promising regions. We evaluate our method on three kinodynamic systems, including physical robot experiments. Results across all three robots tested indicate that RL-RRT outperforms state of the art kinodynamic planners in efficiency, and also provides a shorter path finish time than a steering function free method. The learned local planner policy and accompanying reachability estimator demonstrate transferability to the previously unseen experimental environments, making RL-RRT fast because the expensive computations are replaced with simple neural network inference. Video: https://youtu.be/dDMVMTOI8KY

Modern Reinforcement Learning (RL) is commonly applied to practical problems with an enormous number of states, where function approximation must be deployed to approximate either the value function or the policy. The introduction of function approximation raises a fundamental set of challenges involving computational and statistical efficiency, especially given the need to manage the exploration/exploitation tradeoff. As a result, a core RL question remains open: how can we design provably efficient RL algorithms that incorporate function approximation? This question persists even in a basic setting with linear dynamics and linear rewards, for which only linear function approximation is needed. This paper presents the first provable RL algorithm with both polynomial runtime and polynomial sample complexity in this linear setting, without requiring a "simulator" or additional assumptions. Concretely, we prove that an optimistic modification of Least-Squares Value Iteration (LSVI)---a classical algorithm frequently studied in the linear setting---achieves $\tilde{\mathcal{O}}(\sqrt{d^3H^3T})$ regret, where $d$ is the ambient dimension of feature space, $H$ is the length of each episode, and $T$ is the total number of steps. Importantly, such regret is independent of the number of states and actions.

Many real-world human behaviors can be characterized as a sequential decision making processes, such as urban travelers choices of transport modes and routes (Wu et al. 2017). Differing from choices controlled by machines, which in general follows perfect rationality to adopt the policy with the highest reward, studies have revealed that human agents make sub-optimal decisions under bounded rationality (Tao, Rohde, and Corcoran 2014). Such behaviors can be modeled using maximum causal entropy (MCE) principle (Ziebart 2010). In this paper, we define and investigate a general reward trans-formation problem (namely, reward advancement): Recovering the range of additional reward functions that transform the agent's policy from original policy to a predefined target policy under MCE principle. We show that given an MDP and a target policy, there are infinite many additional reward functions that can achieve the desired policy transformation. Moreover, we propose an algorithm to further extract the additional rewards with minimum "cost" to implement the policy transformation.

Cooperative game is a critical research area in multi-agent reinforcement learning (MARL). Global reward game is a subclass of cooperative games, where all agents aim to maximize cumulative global rewards. Credit assignment is an important problem studied in the global reward game. Most works stand by the view of non-cooperative-game theoretical framework with the shared reward approach, i.e., each agent is assigned a shared global reward directly. This, however, may give each agent an inaccurate feedback on his contribution to the group. In this paper, we introduce a cooperative-game theoretical framework and extend it to the finite-horizon case. We show that our proposed framework is a superset of the global reward game. Based on this framework, we propose an algorithm called Shapley Q-value policy gradient (SQPG) to learn a local reward approach that can distribute the cumulative global reward fairly, reflecting each agent's own contribution in contrast to the shared reward approach. We evaluate our method on the Cooperative Navigation, Prey-and-Predator and Traffic Junction, compared with MADDPG, COMA, Independent actor-critic and Independent DDPG. In the experiments, our algorithm shows better convergence than the baselines.

In multi-task reinforcement learning there are two main challenges: at training time, the ability to learn different policies with a single model; at test time, inferring which of those policies applying without an external signal. In the case of continual reinforcement learning a third challenge arises: learning tasks sequentially without forgetting the previous ones. In this paper, we tackle these challenges by proposing DisCoRL, an approach combining state representation learning and policy distillation. We experiment on a sequence of three simulated 2D navigation tasks with a 3 wheel omni-directional robot. Moreover, we tested our approach's robustness by transferring the final policy into a real life setting. The policy can solve all tasks and automatically infer which one to run.

Batch Reinforcement Learning (Batch RL) consists in training a policy using trajectories collected with another policy, called the behavioural policy. Safe policy improvement (SPI) provides guarantees with high probability that the trained policy performs better than the behavioural policy, also called baseline in this setting. Previous work shows that the SPI objective improves mean performance as compared to using the basic RL objective, which boils down to solving the MDP with maximum likelihood. Here, we build on that work and improve more precisely the SPI with Baseline Bootstrapping algorithm (SPIBB) by allowing the policy search over a wider set of policies. Instead of binarily classifying the state-action pairs into two sets (the \textit{uncertain} and the \textit{safe-to-train-on} ones), we adopt a softer strategy that controls the error in the value estimates by constraining the policy change according to the local model uncertainty. The method can take more risks on uncertain actions all the while remaining provably-safe, and is therefore less conservative than the state-of-the-art methods. We propose two algorithms (one optimal and one approximate) to solve this constrained optimization problem and empirically show a significant improvement over existing SPI algorithms both on finite MDPs and on infinite MDPs with a neural network function approximation.

In instruction conditioned navigation, agents interpret natural language and their surroundings to navigate through an environment. Datasets for studying this task typically contain pairs of these instructions and reference trajectories. Yet, most evaluation metrics used thus far fail to properly account for the latter, relying instead on insufficient similarity comparisons. We address fundamental flaws in previously used metrics and show how Dynamic Time Warping (DTW), a long known method of measuring similarity between two time series, can be used for evaluation of navigation agents. For such, we define the normalized Dynamic Time Warping (nDTW) metric, that softly penalizes deviations from the reference path, is naturally sensitive to the order of the nodes composing each path, is suited for both continuous and graph-based evaluations, and can be efficiently calculated. Further, we define SDTW, which constrains nDTW to only successful paths. We collect human similarity judgments for simulated paths and find nDTW correlates better with human rankings than all other metrics. We also demonstrate that using nDTW as a reward signal for Reinforcement Learning navigation agents improves their performance on both the Room-to-Room (R2R) and Room-for-Room (R4R) datasets. The R4R results in particular highlight the superiority of SDTW over previous success-constrained metrics.

We present Imitation-Projected Policy Gradient (IPPG), an algorithmic framework for learning policies that are parsimoniously represented in a structured programming language. Such programmatic policies can be more interpretable, generalizable, and amenable to formal verification than neural policies; however, designing rigorous learning approaches for programmatic policies remains a challenge. IPPG, our response to this challenge, is based on three insights. First, we view our learning task as optimization in policy space, modulo the constraint that the desired policy has a programmatic representation, and solve this optimization problem using a "lift-and-project" perspective that takes a gradient step into the unconstrained policy space and then projects back onto the constrained space. Second, we view the unconstrained policy space as mixing neural and programmatic representations, which enables employing state-of-the-art deep policy gradient approaches. Third, we cast the projection step as program synthesis via imitation learning, and exploit contemporary combinatorial methods for this task. We present theoretical convergence results for IPPG, as well as an empirical evaluation in three continuous control domains. The experiments show that IPPG can significantly outperform state-of-the-art approaches for learning programmatic policies.