Large Language Models (LLMs) have been shown to act like planners that can decompose high-level instructions into a sequence of executable instructions. However, current LLM-based planners are only able to operate with a fixed set of skills. We overcome this critical limitation and present a method for using LLM-based planners to query new skills and teach robots these skills in a data and time-efficient manner for rigid object manipulation. Our system can re-use newly acquired skills for future tasks, demonstrating the potential of open world and lifelong learning. We evaluate the proposed framework on multiple tasks in simulation and the real world. Videos are available at: https://sites.google.com/mit.edu/halp-robot-learning.

Motion planning for autonomous robots in tight, interaction-rich, and mixed human-robot environments is challenging. State-of-the-art methods typically separate prediction and planning, predicting other agents' trajectories first and then planning the ego agent's motion in the remaining free space. However, agents' lack of awareness of their influence on others can lead to the freezing robot problem. We build upon Interaction-Aware Model Predictive Path Integral (IA-MPPI) control and combine it with learning-based trajectory predictions, thereby relaxing its reliance on communicated short-term goals for other agents. We apply this framework to Autonomous Surface Vessels (ASVs) navigating urban canals. By generating an artificial dataset in real sections of Amsterdam's canals, adapting and training a prediction model for our domain, and proposing heuristics to extract local goals, we enable effective cooperation in planning. Our approach improves autonomous robot navigation in complex, crowded environments, with potential implications for multi-agent systems and human-robot interaction.

In recent years, the need for resources for handling processes with high computational complexity for mobile robots is becoming increasingly urgent. More specifically, robots need to autonomously operate in a robust and continuous manner, while keeping high performance, a need that led to the utilization of edge computing to offload many computationally demanding and time-critical robotic procedures. However, safe mechanisms should be implemented to handle situations when it is not possible to use the offloaded procedures, such as if the communication is challenged or the edge cluster is not available. To this end, this article presents a switching strategy for safety, redundancy, and optimized behavior through an edge computing-based Model Predictive Controller (MPC) and a low-level onboard-PID controller for edge-connected Unmanned Aerial Vehicles (UAVs). The switching strategy is based on the communication Key Performance Indicators (KPIs) over 5G to decide whether the UAV should be controlled by the edge-based or have a safe fallback based on the onboard controller.

Reinforcement Learning (RL) is notoriously data-inefficient, which makes training on a real robot difficult. While model-based RL algorithms (world models) improve data-efficiency to some extent, they still require hours or days of interaction to learn skills. Recently, offline RL has been proposed as a framework for training RL policies on pre-existing datasets without any online interaction. However, constraining an algorithm to a fixed dataset induces a state-action distribution shift between training and inference, and limits its applicability to new tasks. In this work, we seek to get the best of both worlds: we consider the problem of pretraining a world model with offline data collected on a real robot, and then finetuning the model on online data collected by planning with the learned model. To mitigate extrapolation errors during online interaction, we propose to regularize the planner at test-time by balancing estimated returns and (epistemic) model uncertainty. We evaluate our method on a variety of visuo-motor control tasks in simulation and on a real robot, and find that our method enables few-shot finetuning to seen and unseen tasks even when offline data is limited. Videos, code, and data are available at https://yunhaifeng.com/FOWM .

We demonstrate the versatility of mean-field games (MFGs) as a mathematical framework for explaining, enhancing, and designing generative models. In generative flows, a Lagrangian formulation is used where each particle (generated sample) aims to minimize a loss function over its simulated path. The loss, however, is dependent on the paths of other particles, which leads to a competition among the population of particles. The asymptotic behavior of this competition yields a mean-field game. We establish connections between MFGs and major classes of generative flows and diffusions including continuous-time normalizing flows, score-based generative models (SGM), and Wasserstein gradient flows. Furthermore, we study the mathematical properties of each generative model by studying their associated MFG's optimality condition, which is a set of coupled forward-backward nonlinear partial differential equations. The mathematical structure described by the MFG optimality conditions identifies the inductive biases of generative flows. We investigate the well-posedness and structure of normalizing flows, unravel the mathematical structure of SGMs, and derive a MFG formulation of Wasserstein gradient flows. From an algorithmic perspective, the optimality conditions yields Hamilton-Jacobi-Bellman (HJB) regularizers for enhanced training of generative models. In particular, we propose and demonstrate an HJB-regularized SGM with improved performance over standard SGMs. We present this framework as an MFG laboratory which serves as a platform for revealing new avenues of experimentation and invention of generative models.

In a typical stochastic multi-armed bandit problem, the objective is often to maximize the expected sum of rewards over some time horizon $T$. While the choice of a strategy that accomplishes that is optimal with no additional information, it is no longer the case when provided additional environment-specific knowledge. In particular, in areas of high volatility like healthcare or finance, a naive reward maximization approach often does not accurately capture the complexity of the learning problem and results in unreliable solutions. To tackle problems of this nature, we propose a framework of adaptive risk-aware strategies that operate in non-stationary environments. Our framework incorporates various risk measures prevalent in the literature to map multiple families of multi-armed bandit algorithms into a risk-sensitive setting. In addition, we equip the resulting algorithms with the Restarted Bayesian Online Change-Point Detection (R-BOCPD) algorithm and impose a (tunable) forced exploration strategy to detect local (per-arm) switches. We provide finite-time theoretical guarantees and an asymptotic regret bound of order $\tilde O(\sqrt{K_T T})$ up to time horizon $T$ with $K_T$ the total number of change-points. In practice, our framework compares favorably to the state-of-the-art in both synthetic and real-world environments and manages to perform efficiently with respect to both risk-sensitivity and non-stationarity.

Orbital-free density functional theory (OF-DFT) holds the promise to compute ground state molecular properties at minimal cost. However, it has been held back by our inability to compute the kinetic energy as a functional of the electron density only. We here set out to learn the kinetic energy functional from ground truth provided by the more expensive Kohn-Sham density functional theory. Such learning is confronted with two key challenges: Giving the model sufficient expressivity and spatial context while limiting the memory footprint to afford computations on a GPU; and creating a sufficiently broad distribution of training data to enable iterative density optimization even when starting from a poor initial guess. In response, we introduce KineticNet, an equivariant deep neural network architecture based on point convolutions adapted to the prediction of quantities on molecular quadrature grids. Important contributions include convolution filters with sufficient spatial resolution in the vicinity of the nuclear cusp, an atom-centric sparse but expressive architecture that relays information across multiple bond lengths; and a new strategy to generate varied training data by finding ground state densities in the face of perturbations by a random external potential. KineticNet achieves, for the first time, chemical accuracy of the learned functionals across input densities and geometries of tiny molecules. For two electron systems, we additionally demonstrate OF-DFT density optimization with chemical accuracy.

Scientific researchers frequently use the in situ synchrotron high-energy powder X-ray diffraction (XRD) technique to examine the crystallographic structures of materials in functional devices such as rechargeable battery materials. We propose a method for identifying artifacts in experimental XRD images. The proposed method uses deep learning convolutional neural network architectures, such as tunable U-Nets to identify the artifacts. In particular, the predicted artifacts are evaluated against the corresponding ground truth (manually implemented) using the overall true positive rate or recall. The result demonstrates that the U-Nets can consistently produce great recall performance at 92.4% on the test dataset, which is not included in the training, with a 34% reduction in average false positives in comparison to the conventional method. The U-Nets also reduce the time required to identify and separate artifacts by more than 50%. Furthermore, the exclusion of the artifacts shows major changes in the integrated 1D XRD pattern, enhancing further analysis of the post-processing XRD data.

The modern power grid is facing increasing complexities, primarily stemming from the integration of renewable energy sources and evolving consumption patterns. This paper introduces an innovative methodology that harnesses Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) networks to establish robust time series forecasting models for grid frequency. These models effectively capture the spatiotemporal intricacies inherent in grid frequency data, significantly enhancing prediction accuracy and bolstering power grid reliability. The research explores the potential and development of individualized Convolutional LSTM (ConvLSTM) models for buildings within a university campus, enabling them to be independently trained and evaluated for each building. Individual ConvLSTM models are trained on power consumption data for each campus building and forecast the grid frequency based on historical trends. The results convincingly demonstrate the superiority of the proposed models over traditional forecasting techniques, as evidenced by performance metrics such as Mean Square Error (MSE), Mean Absolute Error (MAE), and Mean Absolute Percentage Error (MAPE). Additionally, an Ensemble Model is formulated to aggregate insights from the building-specific models, delivering comprehensive forecasts for the entire campus. This approach ensures the privacy and security of power consumption data specific to each building.

Simulation of autonomous vehicle systems requires that simulated traffic participants exhibit diverse and realistic behaviors. The use of prerecorded real-world traffic scenarios in simulation ensures realism but the rarity of safety critical events makes large scale collection of driving scenarios expensive. In this paper, we present DJINN - a diffusion based method of generating traffic scenarios. Our approach jointly diffuses the trajectories of all agents, conditioned on a flexible set of state observations from the past, present, or future. On popular trajectory forecasting datasets, we report state of the art performance on joint trajectory metrics. In addition, we demonstrate how DJINN flexibly enables direct test-time sampling from a variety of valuable conditional distributions including goal-based sampling, behavior-class sampling, and scenario editing.