The purpose of multi-task reinforcement learning (MTRL) is to train a single policy that can be applied to a set of different tasks. Sharing parameters allows us to take advantage of the similarities among tasks. However, the gaps between contents and difficulties of different tasks bring us challenges on both which tasks should share the parameters and what parameters should be shared, as well as the optimization challenges due to parameter sharing. In this work, we introduce a parameter-compositional approach (PaCo) as an attempt to address these challenges. In this framework, a policy subspace represented by a set of parameters is learned. Policies for all the single tasks lie in this subspace and can be composed by interpolating with the learned set. It allows not only flexible parameter sharing but also a natural way to improve training. We demonstrate the state-of-the-art performance on Meta-World benchmarks, verifying the effectiveness of the proposed approach.

The number of states in a dynamic process is exponential in the number of objects, making reinforcement learning (RL) difficult in complex, multi-object domains. For agents to scale to the real world, they will need to react to and reason about unseen combinations of objects. We argue that the ability to recognize and use local factorization in transition dynamics is a key element in unlocking the power of multi-object reasoning. To this end, we show that (1) known local structure in the environment transitions is sufficient for an exponential reduction in the sample complexity of training a dynamics model, and (2) a locally factored dynamics model provably generalizes out-of-distribution to unseen states and actions. Knowing the local structure also allows us to predict which unseen states and actions this dynamics model will generalize to. We propose to leverage these observations in a novel Model-based Counterfactual Data Augmentation (MoCoDA) framework. MoCoDA applies a learned locally factored dynamics model to an augmented distribution of states and actions to generate counterfactual transitions for RL. MoCoDA works with a broader set of local structures than prior work and allows for direct control over the augmented training distribution. We show that MoCoDA enables RL agents to learn policies that generalize to unseen states and actions. We use MoCoDA to train an offline RL agent to solve an out-of-distribution robotics manipulation task on which standard offline RL algorithms fail.

In this paper, we present a learning-based approach that allows a robot to quickly follow a reference path defined in joint space without exceeding limits on the position, velocity, acceleration and jerk of each robot joint. Contrary to offline methods for time-optimal path parameterization, the reference path can be changed during motion execution. In addition, our approach can utilize sensory feedback, for instance, to follow a reference path with a bipedal robot without losing balance. With our method, the robot is controlled by a neural network that is trained via reinforcement learning using data generated by a physics simulator. From a mathematical perspective, the problem of tracking a reference path in a time-optimized manner is formalized as a Markov decision process. Each state includes a fixed number of waypoints specifying the next part of the reference path. The action space is designed in such a way that all resulting motions comply with the specified kinematic joint limits. The reward function finally reflects the trade-off between the execution time, the deviation from the desired reference path and optional additional objectives like balancing. We evaluate our approach with and without additional objectives and show that time-optimized path tracking can be successfully learned for both industrial and humanoid robots. In addition, we demonstrate that networks trained in simulation can be successfully transferred to a real robot.

Imitation learning aims to extract knowledge from human experts' demonstrations or artificially created agents in order to replicate their behaviors. Its success has been demonstrated in areas such as video games, autonomous driving, robotic simulations and object manipulation. However, this replicating process could be problematic, such as the performance is highly dependent on the demonstration quality, and most trained agents are limited to perform well in task-specific environments. In this survey, we provide a systematic review on imitation learning. We first introduce the background knowledge from development history and preliminaries, followed by presenting different taxonomies within Imitation Learning and key milestones of the field. We then detail challenges in learning strategies and present research opportunities with learning policy from suboptimal demonstration, voice instructions and other associated optimization schemes.

Training a reinforcement learning (RL) agent on a real-world robotics task remains generally impractical due to sample inefficiency. Multi-task RL and meta-RL aim to improve sample efficiency by generalizing over a distribution of related tasks. However, doing so is difficult in practice: In multi-task RL, state of the art methods often fail to outperform a degenerate solution that simply learns each task separately. Hypernetworks are a promising path forward since they replicate the separate policies of the degenerate solution while also allowing for generalization across tasks, and are applicable to meta-RL. However, evidence from supervised learning suggests hypernetwork performance is highly sensitive to the initialization. In this paper, we 1) show that hypernetwork initialization is also a critical factor in meta-RL, and that naive initializations yield poor performance; 2) propose a novel hypernetwork initialization scheme that matches or exceeds the performance of a state-of-the-art approach proposed for supervised settings, as well as being simpler and more general; and 3) use this method to show that hypernetworks can improve performance in meta-RL by evaluating on multiple simulated robotics benchmarks.

We identify and study the phenomenon of policy churn, that is, the rapid change of the greedy policy in value-based reinforcement learning. Policy churn operates at a surprisingly rapid pace, changing the greedy action in a large fraction of states within a handful of learning updates (in a typical deep RL set-up such as DQN on Atari). We characterise the phenomenon empirically, verifying that it is not limited to specific algorithm or environment properties. A number of ablations help whittle down the plausible explanations on why churn occurs to just a handful, all related to deep learning. Finally, we hypothesise that policy churn is a beneficial but overlooked form of implicit exploration that casts $\epsilon$-greedy exploration in a fresh light, namely that $\epsilon$-noise plays a much smaller role than expected.

A cognitive radar is a constrained utility maximizer that adapts its sensing mode in response to a changing environment. If an adversary can estimate the utility function of a cognitive radar, it can determine the radar's sensing strategy and mitigate the radar performance via electronic countermeasures (ECM). This paper discusses how a cognitive radar can {\em hide} its strategy from an adversary that detects cognition. The radar does so by transmitting purposefully designed sub-optimal responses to spoof the adversary's Neyman-Pearson detector. We provide theoretical guarantees by ensuring the Type-I error probability of the adversary's detector exceeds a pre-defined level for a specified tolerance on the radar's performance loss. We illustrate our cognition masking scheme via numerical examples involving waveform adaptation and beam allocation. We show that small purposeful deviations from the optimal strategy of the radar confuse the adversary by significant amounts, thereby masking the radar's cognition. Our approach uses novel ideas from revealed preference in microeconomics and adversarial inverse reinforcement learning. Our proposed algorithms provide a principled approach for system-level electronic counter-countermeasures (ECCM) to mask the radar's cognition, i.e., hide the radar's strategy from an adversary. We also provide performance bounds for our cognition masking scheme when the adversary has misspecified measurements of the radar's response.

Designing reinforcement learning (RL) agents is typically a difficult process that requires numerous design iterations. Learning can fail for a multitude of reasons, and standard RL methods provide too few tools to provide insight into the exact cause. In this paper, we show how to integrate value decomposition into a broad class of actor-critic algorithms and use it to assist in the iterative agent-design process. Value decomposition separates a reward function into distinct components and learns value estimates for each. These value estimates provide insight into an agent's learning and decision-making process and enable new training methods to mitigate common problems. As a demonstration, we introduce SAC-D, a variant of soft actor-critic (SAC) adapted for value decomposition. SAC-D maintains similar performance to SAC, while learning a larger set of value predictions. We also introduce decomposition-based tools that exploit this information, including a new reward influence metric, which measures each reward component's effect on agent decision-making. Using these tools, we provide several demonstrations of decomposition's use in identifying and addressing problems in the design of both environments and agents. Value decomposition is broadly applicable and easy to incorporate into existing algorithms and workflows, making it a powerful tool in an RL practitioner's toolbox.

This study addresses the problem of selecting dynamically, at each time instance, the ``optimal'' p-norm to combat outliers in linear adaptive filtering without any knowledge on the potentially time-varying probability distribution function of the outliers. To this end, an online and data-driven framework is designed via kernel-based reinforcement learning (KBRL). Novel Bellman mappings on reproducing kernel Hilbert spaces (RKHSs) are introduced that need no knowledge on transition probabilities of Markov decision processes, and are nonexpansive with respect to the underlying Hilbertian norm. An approximate policy-iteration framework is finally offered via the introduction of a finite-dimensional affine superset of the fixed-point set of the proposed Bellman mappings. The well-known ``curse of dimensionality'' in RKHSs is addressed by building a basis of vectors via an approximate linear dependency criterion. Numerical tests on synthetic data demonstrate that the proposed framework selects always the ``optimal'' p-norm for the outlier scenario at hand, outperforming at the same time several non-RL and KBRL schemes.

Autonomous navigation in dynamic environments is a complex but essential task for autonomous robots. Recent deep reinforcement learning approaches show promising results to solve the problem, but it is not solved yet, as they typically assume no robot kinodynamic restrictions, holonomic movement or perfect environment knowledge. Moreover, most algorithms fail in the real world due to the inability to generate real-world training data for the huge variability of possible scenarios. In this work, we present a novel planner, DQN-DOVS, that uses deep reinforcement learning on a descriptive robocentric velocity space model to navigate in highly dynamic environments. It is trained using a smart curriculum learning approach on a simulator that faithfully reproduces the real world, reducing the gap between the reality and simulation. We test the resulting algorithm in scenarios with different number of obstacles and compare it with many state-of-the-art approaches, obtaining a better performance. Finally, we try the algorithm in a ground robot, using the same setup as in the simulation experiments.