Open-set task execution can significantly benefit from seamlessly switching between coarse and fine scene representations depending on the context and the evolving information as the robot explores the environment. For example, it is often sufficient to start with a coarse scene representation initially and only employ a finer, more granular scene representation when the robot encounters regions which are likely to contain the task relevant objects. Hence, in this work, we propose BiMoSG, a bimodal 3D scene graph generation approach for open-set tasks. BiMoSG employs a "fast" mode by default to efficiently generate a coarse 3D scene graph and can switch to a "slow" mode for generating a finer open vocabulary 3D scene graph of task relevant objects. We demonstrate that our proposed 3D scene graph generation approach is significantly faster than the open-source state-of-the-art approaches. This allows us to integrate the scene graph generation process with task execution for real-time deployment.

Multi-agent pursuit-evasion scenarios in the presence of learning targets pose a significant challenge for robot swarms. In this work, we investigate a cooperative strategy among decentralized unmanned aerial vehicles to keep track of a rapidly evading target afforded with learning capabilities. To deal with the non-stationarity of the environment while solely depending on local observations, we propose a multi-agent reinforcement learning framework enhanced with dynamic intraswarm communication protocols available during the training process. Specifically, we propose, (a) a coverage-range-based communication network, and (b) a $k$-nearest neighborhood communication network which are both developed based on reward policies aimed at maintaining network connectivity despite its dynamic nature. Both of these protocols are integrated into the training process as additional reward-shaping terms, which allow the agents to be deployed in a decentralized manner without requiring any information exchange based on a learned joint policy that adapts to the presence of a communication channel. The multi-agent system is trained in a simulated environment using the Multiple Particle Environment and deployed on Crazyflie drones in a real-world controlled environment. Our experimental results in simulation and real-world show that swarms trained without our proposed networking protocols fail to learn any effective policy. In addition, our proposed dynamic communication strategy during training yields two notable benefits to the system's performance during deployment: (i) it reduces the average distance to the target agent, and (ii) it enables the learning of the pursuit task in larger environments.

A major bottleneck in the training process for Zero-Shot Coordination (ZSC) agents is the generation of partner agents that are diverse in collaborative conventions. Current Cross-play Minimization (XPM) methods for population generation can be very computationally expensive and sample inefficient as the training objective requires sampling multiple types of trajectories. Each partner agent in the population is also trained from scratch, despite all of the partners in the population learning policies of the same coordination task. In this work, we propose that simulated trajectories from the dynamics model of an environment can drastically speed up the training process for XPM methods. We introduce XPM-WM, a framework for generating simulated trajectories for XPM via a learned World Model (WM). We show XPM with simulated trajectories removes the need to sample multiple trajectories. In addition, we show our proposed method can effectively generate partners with diverse conventions that match the performance of previous methods in terms of SP population training reward as well as training partners for ZSC agents. Our method is thus, significantly more sample efficient and scalable to a larger number of partners.

A major challenge for deep reinforcement learning (DRL) agents is to collaborate with novel partners that were not encountered by them during the training phase. This is specifically worsened by an increased variance in action responses when the DRL agents collaborate with human partners due to the lack of consistency in human behaviors. Recent work have shown that training a single agent as the best response to a diverse population of training partners significantly increases an agent's robustness to novel partners. We further enhance the population-based training approach by introducing a Hierarchical Reinforcement Learning (HRL) based method for Human-AI Collaboration. Our agent is able to learn multiple best-response policies as its low-level policy while at the same time, it learns a high-level policy that acts as a manager which allows the agent to dynamically switch between the low-level best-response policies based on its current partner. We demonstrate that our method is able to dynamically adapt to novel partners of different play styles and skill levels in the 2-player collaborative Overcooked game environment. We also conducted a human study in the same environment to test the effectiveness of our method when partnering with real human subjects. Code is available at https://gitlab.com/marvl-hipt/hipt.

Recent few-shot learning algorithms have enabled models to quickly adapt to new tasks based on only a few training samples. Previous few-shot learning works have mainly focused on classification and reinforcement learning. In this paper, we propose a few-shot meta-learning system that focuses exclusively on regression tasks. Our model is based on the idea that the degree of freedom of the unknown function can be significantly reduced if it is represented as a linear combination of a set of sparsifying basis functions. This enables a few labeled samples to approximate the function. We design a Basis Function Learner network to encode basis functions for a task distribution, and a Weights Generator network to generate the weight vector for a novel task. We show that our model outperforms the current state of the art meta-learning methods in various regression tasks.

The recent rise in popularity of few-shot learning algorithms has enabled models to quickly adapt to new tasks based on only a few training samples. Previous few-shot learning works have mainly focused on classification and reinforcement learning. In this paper, we propose a few-shot meta learning system that focuses exclusively on regression tasks. Our model is based on the idea that the degree of freedom of the unknown function can be significantly reduced if it is represented as a linear combination of a set of appropriate basis functions. This enables a few labelled samples to approximate the function. We design a Feature Extractor network to encode basis functions for a task distribution, and a Weights Generator to generate the weight vector for a novel task. We show that our model outperforms the current state of the art meta-learning methods in various regression tasks.

Recently, the introduction of the generative adversarial network (GAN) and its variants has enabled the generation of realistic synthetic samples, which has been used for enlarging training sets. Previous work primarily focused on data augmentation for semi-supervised and supervised tasks. In this paper, we instead focus on unsupervised anomaly detection and propose a novel generative data augmentation framework optimized for this task. In particular, we propose to oversample infrequent normal samples - normal samples that occur with small probability, e.g., rare normal events. We show that these samples are responsible for false positives in anomaly detection. However, oversampling of infrequent normal samples is challenging for real-world high-dimensional data with multimodal distributions. To address this challenge, we propose to use a GAN variant known as the adversarial autoencoder (AAE) to transform the high-dimensional multimodal data distributions into low-dimensional unimodal latent distributions with well-defined tail probability. Then, we systematically oversample at the `edge' of the latent distributions to increase the density of infrequent normal samples. We show that our oversampling pipeline is a unified one: it is generally applicable to datasets with different complex data distributions. To the best of our knowledge, our method is the first data augmentation technique focused on improving performance in unsupervised anomaly detection. We validate our method by demonstrating consistent improvements across several real-world datasets.