Robot learning requires a considerable amount of high-quality data to realize the promise of generalization. However, large data sets are costly to collect in the real world. Physics simulators can cheaply generate vast data sets with broad coverage over states, actions, and environments. However, physics engines are fundamentally misspecified approximations to reality. This makes direct zero-shot transfer from simulation to reality challenging, especially in tasks where precise and force-sensitive manipulation is necessary. Thus, fine-tuning these policies with small real-world data sets is an appealing pathway for scaling robot learning. However, current reinforcement learning fine-tuning frameworks leverage general, unstructured exploration strategies which are too inefficient to make real-world adaptation practical. This paper introduces the Simulation-Guided Fine-tuning (SGFT) framework, which demonstrates how to extract structural priors from physics simulators to substantially accelerate real-world adaptation. Specifically, our approach uses a value function learned in simulation to guide real-world exploration. We demonstrate this approach across five real-world dexterous manipulation tasks where zero-shot sim-to-real transfer fails. We further demonstrate our framework substantially outperforms baseline fine-tuning methods, requiring up to an order of magnitude fewer real-world samples and succeeding at difficult tasks where prior approaches fail entirely. Last but not least, we provide theoretical justification for this new paradigm which underpins how SGFT can rapidly learn high-performance policies in the face of large sim-to-real dynamics gaps. Project webpage: https://weirdlabuw.github.io/sgft/{weirdlabuw.github.io/sgft}

In order to mitigate the sample complexity of real-world reinforcement learning, common practice is to first train a policy in a simulator where samples are cheap, and then deploy this policy in the real world, with the hope that it generalizes effectively. Such \emph{direct sim2real} transfer is not guaranteed to succeed, however, and in cases where it fails, it is unclear how to best utilize the simulator. In this work, we show that in many regimes, while direct sim2real transfer may fail, we can utilize the simulator to learn a set of \emph{exploratory} policies which enable efficient exploration in the real world. In particular, in the setting of low-rank MDPs, we show that coupling these exploratory policies with simple, practical approaches -- least-squares regression oracles and naive randomized exploration -- yields a polynomial sample complexity in the real world, an exponential improvement over direct sim2real transfer, or learning without access to a simulator. To the best of our knowledge, this is the first evidence that simulation transfer yields a provable gain in reinforcement learning in settings where direct sim2real transfer fails. We validate our theoretical results on several realistic robotic simulators and a real-world robotic sim2real task, demonstrating that transferring exploratory policies can yield substantial gains in practice as well.

Speech foundation models, trained on vast datasets, have opened unique opportunities in addressing challenging low-resource speech understanding, such as child speech. In this work, we explore the capabilities of speech foundation models on child-adult speaker diarization. We show that exemplary foundation models can achieve 39.5% and 62.3% relative reductions in Diarization Error Rate and Speaker Confusion Rate, respectively, compared to previous speaker diarization methods. In addition, we benchmark and evaluate the speaker diarization results of the speech foundation models with varying the input audio window size, speaker demographics, and training data ratio. Our results highlight promising pathways for understanding and adopting speech foundation models to facilitate child speech understanding.

Executing precise and agile flight maneuvers is important for the ongoing commoditization of unmanned aerial vehicles (UAVs), in applications such as drone delivery, rescue and search, and urban air mobility. In particular, accurately following arbitrary trajectories with quadrotors is among the most notable challenges to precise flight control for the following reasons. First, quadrotor dynamics are highly nonlinear and underactuated, and often hard to model due to unknown system parameters (e.g., motor characteristics) and uncertain environments (e.g., complex aerodynamics from unknown wind gusts). Second, aggressive trajectories demand operating at the limits of system performance, requiring awareness and proper handling of actuation constraints, especially for quadrotors with small thrust-to-weight ratios. Finally, the arbitrary desired trajectory might not be dynamically feasible (i.e., impossible to stay on such a trajectory), which necessities long-horizon reasoning and optimization in real-time. For instance, to stay close to the five-star trajectory in Figure 1, which is infeasible due to the sharp changes of direction, the quadrotor must predict, plan, and react online before the sharp turns.

A major challenge in robotics is to design robust policies which enable complex and agile behaviors in the real world. On one end of the spectrum, we have model-free reinforcement learning (MFRL), which is incredibly flexible and general but often results in brittle policies. In contrast, model predictive control (MPC) continually re-plans at each time step to remain robust to perturbations and model inaccuracies. However, despite its real-world successes, MPC often under-performs the optimal strategy. This is due to model quality, myopic behavior from short planning horizons, and approximations due to computational constraints. And even with a perfect model and enough compute, MPC can get stuck in bad local optima, depending heavily on the quality of the optimization algorithm. To this end, we propose Deep Model Predictive Optimization (DMPO), which learns the inner-loop of an MPC optimization algorithm directly via experience, specifically tailored to the needs of the control problem. We evaluate DMPO on a real quadrotor agile trajectory tracking task, on which it improves performance over a baseline MPC algorithm for a given computational budget. It can outperform the best MPC algorithm by up to 27% with fewer samples and an end-to-end policy trained with MFRL by 19%. Moreover, because DMPO requires fewer samples, it can also achieve these benefits with 4.3X less memory. When we subject the quadrotor to turbulent wind fields with an attached drag plate, DMPO can adapt zero-shot while still outperforming all baselines. Additional results can be found at https://tinyurl.com/mr2ywmnw.

The architecture of a deep neural network is defined explicitly in terms of the number of layers, the width of each layer and the general network topology. Existing optimisation frameworks neglect this information in favour of implicit architectural information (e.g. second-order methods) or architecture-agnostic distance functions (e.g. mirror descent). Meanwhile, the most popular optimiser in practice, Adam, is based on heuristics. This paper builds a new framework for deriving optimisation algorithms that explicitly leverage neural architecture. The theory extends mirror descent to non-convex composite objective functions: the idea is to transform a Bregman divergence to account for the non-linear structure of neural architecture. Working through the details for deep fully-connected networks yields automatic gradient descent: a first-order optimiser without any hyperparameters. Automatic gradient descent trains both fully-connected and convolutional networks out-of-the-box and at ImageNet scale. A PyTorch implementation is available at https://github.com/jxbz/agd and also in Appendix B. Overall, the paper supplies a rigorous theoretical foundation for a next-generation of architecture-dependent optimisers that work automatically and without hyperparameters.

Exposure to ideas in domains outside a scientist's own may benefit her in reformulating existing research problems in novel ways and discovering new application domains for existing solution ideas. While improved performance in scholarly search engines can help scientists efficiently identify relevant advances in domains they may already be familiar with, it may fall short of helping them explore diverse ideas \textit{outside} such domains. In this paper we explore the design of systems aimed at augmenting the end-user ability in cross-domain exploration with flexible query specification. To this end, we develop an exploratory search system in which end-users can select a portion of text core to their interest from a paper abstract and retrieve papers that have a high similarity to the user-selected core aspect but differ in terms of domains. Furthermore, end-users can `zoom in' to specific domain clusters to retrieve more papers from them and understand nuanced differences within the clusters. Our case studies with scientists uncover opportunities and design implications for systems aimed at facilitating cross-domain exploration and inspiration.

Current state-of-the-art model-based reinforcement learning algorithms use trajectory sampling methods, such as the Cross-Entropy Method (CEM), for planning in continuous control settings. These zeroth-order optimizers require sampling a large number of trajectory rollouts to select an optimal action, which scales poorly for large prediction horizons or high dimensional action spaces. First-order methods that use the gradients of the rewards with respect to the actions as an update can mitigate this issue, but suffer from local optima due to the non-convex optimization landscape. To overcome these issues and achieve the best of both worlds, we propose a novel planner, Cross-Entropy Method with Gradient Descent (CEM-GD), that combines first-order methods with CEM. At the beginning of execution, CEM-GD uses CEM to sample a significant amount of trajectory rollouts to explore the optimization landscape and avoid poor local minima. It then uses the top trajectories as initialization for gradient descent and applies gradient updates to each of these trajectories to find the optimal action sequence. At each subsequent time step, however, CEM-GD samples much fewer trajectories from CEM before applying gradient updates. We show that as the dimensionality of the planning problem increases, CEM-GD maintains desirable performance with a constant small number of samples by using the gradient information, while avoiding local optima using initially well-sampled trajectories. Furthermore, CEM-GD achieves better performance than CEM on a variety of continuous control benchmarks in MuJoCo with 100x fewer samples per time step, resulting in around 25% less computation time and 10% less memory usage. The implementation of CEM-GD is available at $\href{https://github.com/KevinHuang8/CEM-GD}{\text{https://github.com/KevinHuang8/CEM-GD}}$.

Having a comprehensive, high-quality dataset of road sign annotation is critical to the success of AI-based Road Sign Recognition (RSR) systems. In practice, annotators often face difficulties in learning road sign systems of different countries; hence, the tasks are often time-consuming and produce poor results. We propose a novel approach using knowledge graphs and a machine learning algorithm - variational prototyping-encoder (VPE) - to assist human annotators in classifying road signs effectively. Annotators can query the Road Sign Knowledge Graph using visual attributes and receive closest matching candidates suggested by the VPE model. The VPE model uses the candidates from the knowledge graph and a real sign image patch as inputs. We show that our knowledge graph approach can reduce sign search space by 98.9%. Furthermore, with VPE, our system can propose the correct single candidate for 75% of signs in the tested datasets, eliminating the human search effort entirely in those cases.

In Artificial Intelligence we often seek to identify an unknown target function of many variables $y=f(\mathbf{x})$ giving a limited set of instances $S=\{(\mathbf{x^{(i)}},y^{(i)})\}$ with $\mathbf{x^{(i)}} \in D$ where $D$ is a domain of interest. We refer to $S$ as the training set and the final quest is to identify the mathematical model that approximates this target function for new $\mathbf{x}$; with the set $T=\{ \mathbf{x^{(j)}} \} \subset D$ with $T \neq S$ (i.e. thus testing the model generalisation). However, for some applications, the main interest is approximating well the unknown function on a larger domain $D'$ that contains $D$. In cases involving the design of new structures, for instance, we may be interested in maximizing $f$; thus, the model derived from $S$ alone should also generalize well in $D'$ for samples with values of $y$ larger than the largest observed in $S$. In that sense, the AI system would provide important information that could guide the design process, e.g., using the learned model as a surrogate function to design new lab experiments. We introduce a method for multivariate regression based on iterative fitting of a continued fraction by incorporating additive spline models. We compared it with established methods such as AdaBoost, Kernel Ridge, Linear Regression, Lasso Lars, Linear Support Vector Regression, Multi-Layer Perceptrons, Random Forests, Stochastic Gradient Descent and XGBoost. We tested the performance on the important problem of predicting the critical temperature of superconductors based on physical-chemical characteristics.