Teaching an agent to perform new tasks using natural language can easily be hindered by ambiguities in interpretation. When a teacher provides an instruction to a learner about an object by referring to its features, the learner can misunderstand the teacher's intentions, for instance if the instruction ambiguously refer to features of the object, a phenomenon called referential ambiguity. We study how two concepts derived from cognitive sciences can help resolve those referential ambiguities: pedagogy (selecting the right instructions) and pragmatism (learning the preferences of the other agents using inductive reasoning). We apply those ideas to a teacher/learner setup with two artificial agents on a simulated robotic task (block-stacking). We show that these concepts improve sample efficiency for training the learner.

Electric vehicles (EVs) play critical roles in autonomous mobility-on-demand (AMoD) systems, but their unique charging patterns increase the model uncertainties in AMoD systems (e.g. state transition probability). Since there usually exists a mismatch between the training and test/true environments, incorporating model uncertainty into system design is of critical importance in real-world applications. However, model uncertainties have not been considered explicitly in EV AMoD system rebalancing by existing literature yet, and the coexistence of model uncertainties and constraints that the decision should satisfy makes the problem even more challenging. In this work, we design a robust and constrained multi-agent reinforcement learning (MARL) framework with state transition kernel uncertainty for EV AMoD systems. We then propose a robust and constrained MARL algorithm (ROCOMA) with robust natural policy gradients (RNPG) that trains a robust EV rebalancing policy to balance the supply-demand ratio and the charging utilization rate across the city under model uncertainty. Experiments show that the ROCOMA can learn an effective and robust rebalancing policy. It outperforms non-robust MARL methods in the presence of model uncertainties. It increases the system fairness by 19.6% and decreases the rebalancing costs by 75.8%.

This paper provides a finite-sample analysis of a passive stochastic gradient Langevin dynamics algorithm (PSGLD) designed to achieve adaptive inverse reinforcement learning (IRL). By passive, we mean that the noisy gradients available to the PSGLD algorithm (inverse learning process) are evaluated at randomly chosen points by an external stochastic gradient algorithm (forward learner) that aims to optimize a cost function. The PSGLD algorithm acts as a randomized sampler to achieve adaptive IRL by reconstructing this cost function nonparametrically from the stationary measure of a Langevin diffusion. Previous work has analyzed the asymptotic performance of this passive algorithm using weak convergence techniques. This paper analyzes the non-asymptotic (finite-sample) performance using a logarithmic-Sobolev inequality and the Otto-Villani Theorem. We obtain finite-sample bounds on the 2-Wasserstein distance between the estimates generated by the PSGLD algorithm and the cost function. Apart from achieving finite-sample guarantees for adaptive IRL, this work extends a line of research in analysis of passive stochastic gradient algorithms to the finite-sample regime for Langevin dynamics.

In this paper, a novel Multi-agent Reinforcement Learning (MARL) approach, Multi-Agent Continuous Dynamic Policy Gradient (MACDPP) was proposed to tackle the issues of limited capability and sample efficiency in various scenarios controlled by multiple agents. It alleviates the inconsistency of multiple agents' policy updates by introducing the relative entropy regularization to the Centralized Training with Decentralized Execution (CTDE) framework with the Actor-Critic (AC) structure. Evaluated by multi-agent cooperation and competition tasks and traditional control tasks including OpenAI benchmarks and robot arm manipulation, MACDPP demonstrates significant superiority in learning capability and sample efficiency compared with both related multi-agent and widely implemented signal-agent baselines and therefore expands the potential of MARL in effectively learning challenging control scenarios.

Visual Reinforcement Learning (Visual RL), coupled with high-dimensional observations, has consistently confronted the long-standing challenge of out-of-distribution generalization. Despite the focus on algorithms aimed at resolving visual generalization problems, we argue that the devil is in the existing benchmarks as they are restricted to isolated tasks and generalization categories, undermining a comprehensive evaluation of agents' visual generalization capabilities. To bridge this gap, we introduce RL-ViGen: a novel Reinforcement Learning Benchmark for Visual Generalization, which contains diverse tasks and a wide spectrum of generalization types, thereby facilitating the derivation of more reliable conclusions. Furthermore, RL-ViGen incorporates the latest generalization visual RL algorithms into a unified framework, under which the experiment results indicate that no single existing algorithm has prevailed universally across tasks. Our aspiration is that RL-ViGen will serve as a catalyst in this area, and lay a foundation for the future creation of universal visual generalization RL agents suitable for real-world scenarios. Access to our code and implemented algorithms is provided at https://gemcollector.github.io/RL-ViGen/.

In many multi-agent and high-dimensional robotic tasks, the controller can be designed in either a centralized or decentralized way. Correspondingly, it is possible to use either single-agent reinforcement learning (SARL) or multi-agent reinforcement learning (MARL) methods to learn such controllers. However, the relationship between these two paradigms remains under-studied in the literature. This work explores research questions in terms of robustness and performance of SARL and MARL approaches to the same task, in order to gain insight into the most suitable methods. We start by analytically showing the equivalence between these two paradigms under the full-state observation assumption. Then, we identify a broad subclass of \textit{Dec-POMDP} tasks where the agents are weakly or partially interacting. In these tasks, we show that partial observations of each agent are sufficient for near-optimal decision-making. Furthermore, we propose to exploit such partially observable MARL to improve the robustness of robots when joint or agent failures occur. Our experiments on both simulated multi-agent tasks and a real robot task with a mobile manipulator validate the presented insights and the effectiveness of the proposed robust robot learning method via partially observable MARL.

Zero-shot reinforcement learning (RL) promises to provide agents that can perform any task in an environment after an offline pre-training phase. Forward-backward (FB) representations represent remarkable progress towards this ideal, achieving 85% of the performance of task-specific agents in this setting. However, such performance is contingent on access to large and diverse datasets for pre-training, which cannot be expected for most real problems. Here, we explore how FB performance degrades when trained on small datasets that lack diversity, and mitigate it with conservatism, a well-established feature of performant offline RL algorithms. We evaluate our family of methods across various datasets, domains and tasks, reaching 150% of vanilla FB performance in aggregate. Somewhat surprisingly, conservative FB algorithms also outperform the task-specific baseline, despite lacking access to reward labels and being required to maintain policies for all tasks. Conservative FB algorithms perform no worse than FB on full datasets, and so present little downside over their predecessor. Our code is available open-source via https://enjeeneer.io/projects/conservative-world-models/.

In remote sensing imagery analysis, patch-based methods have limitations in capturing information beyond the sliding window. This shortcoming poses a significant challenge in processing complex and variable geo-objects, which results in semantic inconsistency in segmentation results. To address this challenge, we propose a dynamic scale perception framework, named GeoAgent, which adaptively captures appropriate scale context information outside the image patch based on the different geo-objects. In GeoAgent, each image patch's states are represented by a global thumbnail and a location mask. The global thumbnail provides context beyond the patch, and the location mask guides the perceived spatial relationships. The scale-selection actions are performed through a Scale Control Agent (SCA). A feature indexing module is proposed to enhance the ability of the agent to distinguish the current image patch's location. The action switches the patch scale and context branch of a dual-branch segmentation network that extracts and fuses the features of multi-scale patches. The GeoAgent adjusts the network parameters to perform the appropriate scale-selection action based on the reward received for the selected scale. The experimental results, using two publicly available datasets and our newly constructed dataset WUSU, demonstrate that GeoAgent outperforms previous segmentation methods, particularly for large-scale mapping applications.

In public spaces shared with humans, ensuring multi-robot systems navigate without collisions while respecting social norms is challenging, particularly with limited communication. Although current robot social navigation techniques leverage advances in reinforcement learning and deep learning, they frequently overlook robot dynamics in simulations, leading to a simulation-to-reality gap. In this paper, we bridge this gap by presenting a new multi-robot social navigation environment crafted using Dec-POSMDP and multi-agent reinforcement learning. Furthermore, we introduce SAMARL: a novel benchmark for cooperative multi-robot social navigation. SAMARL employs a unique spatial-temporal transformer combined with multi-agent reinforcement learning. This approach effectively captures the complex interactions between robots and humans, thus promoting cooperative tendencies in multi-robot systems. Our extensive experiments reveal that SAMARL outperforms existing baseline and ablation models in our designed environment. Demo videos for this work can be found at: https://sites.google.com/view/samarl

Developing robotic technologies for use in human society requires ensuring the safety of robots' navigation behaviors while adhering to pedestrians' expectations and social norms. However, maintaining real-time communication between robots and pedestrians to avoid collisions can be challenging. To address these challenges, we propose a novel socially-aware navigation benchmark called NaviSTAR, which utilizes a hybrid Spatio-Temporal grAph tRansformer (STAR) to understand interactions in human-rich environments fusing potential crowd multi-modal information. We leverage off-policy reinforcement learning algorithm with preference learning to train a policy and a reward function network with supervisor guidance. Additionally, we design a social score function to evaluate the overall performance of social navigation. To compare, we train and test our algorithm and other state-of-the-art methods in both simulator and real-world scenarios independently. Our results show that NaviSTAR outperforms previous methods with outstanding performance\footnote{The source code and experiment videos of this work are available at: https://sites.google.com/view/san-navistar