Incorporating contrastive learning objectives in sentence representation learning (SRL) has yielded significant improvements on many sentence-level NLP tasks. However, it is not well understood why contrastive learning works for learning sentence-level semantics. In this paper, we aim to help guide future designs of sentence representation learning methods by taking a closer look at contrastive SRL through the lens of isotropy, contextualization and learning dynamics. We interpret its successes through the geometry of the representation shifts and show that contrastive learning brings isotropy, and drives high intra-sentence similarity: when in the same sentence, tokens converge to similar positions in the semantic space. We also find that what we formalize as "spurious contextualization" is mitigated for semantically meaningful tokens, while augmented for functional ones. We find that the embedding space is directed towards the origin during training, with more areas now better defined. We ablate these findings by observing the learning dynamics with different training temperatures, batch sizes and pooling methods.

With the rise of task-specific pre-training objectives, abstractive summarization models like PEGASUS offer appealing zero-shot performance on downstream summarization tasks. However, the performance of such unsupervised models still lags significantly behind their supervised counterparts. Similarly to the supervised setup, we notice a very high variance in quality among summary candidates from these models while only one candidate is kept as the summary output. In this paper, we propose to re-rank summary candidates in an unsupervised manner, aiming to close the performance gap between unsupervised and supervised models. Our approach improves the unsupervised PEGASUS by up to 7.27% and ChatGPT by up to 6.86% relative mean ROUGE across four widely-adopted summarization benchmarks ; and achieves relative gains of 7.51% (up to 23.73% from XSum to WikiHow) averaged over 30 zero-shot transfer setups (finetuning on a dataset, evaluating on another).

Driving under varying road conditions is challenging, especially for autonomous vehicles that must adapt in real-time to changes in the environment, e.g., rain, snow, etc. It is difficult to apply offline learning-based methods in these time-varying settings, as the controller should be trained on datasets representing all conditions it might encounter in the future. While online learning may adapt a model from real-time data, its convergence is often too slow for fast varying road conditions. We study this problem in autonomous racing, where driving at the limits of handling under varying road conditions is required for winning races. We propose a computationally-efficient approach that leverages an ensemble of Gaussian processes (GPs) to generalize and adapt pre-trained GPs to unseen conditions. Each GP is trained on driving data with a different road surface friction. A time-varying convex combination of these GPs is used within a model predictive control (MPC) framework, where the model weights are adapted online to the current road condition based on real-time data. The predictive variance of the ensemble Gaussian process (EGP) model allows the controller to account for prediction uncertainty and enables safe autonomous driving. Extensive simulations of a full scale autonomous car demonstrated the effectiveness of our proposed EGP-MPC method for providing good tracking performance in varying road conditions and the ability to generalize to unknown maps.

Interest in autonomous driving (AD) and intelligent vehicles (IVs) is growing at a rapid pace due to the convenience, safety, and economic benefits. Although a number of surveys have reviewed research achievements in this field, they are still limited in specific tasks and lack systematic summaries and research directions in the future. Our work is divided into 3 independent articles and the first part is a Survey of Surveys (SoS) for total technologies of AD and IVs that involves the history, summarizes the milestones, and provides the perspectives, ethics, and future research directions. This is the second part (Part I for this technical survey) to review the development of control, computing system design, communication, High Definition map (HD map), testing, and human behaviors in IVs. In addition, the third part (Part II for this technical survey) is to review the perception and planning sections. The objective of this paper is to involve all the sections of AD, summarize the latest technical milestones, and guide abecedarians to quickly understand the development of AD and IVs. Combining the SoS and Part II, we anticipate that this work will bring novel and diverse insights to researchers and abecedarians, and serve as a bridge between past and future.

In this paper, we propose the Phy-DRL: a physics-model-regulated deep reinforcement learning framework for safety-critical autonomous systems. The Phy-DRL is unique in three innovations: i) proactive unknown-unknowns training, ii) conjunctive residual control (i.e., integration of data-driven control and physics-model-based control) and safety- \& stability-sensitive reward, and iii) physics-model-based neural network editing, including link editing and activation editing. Thanks to the concurrent designs, the Phy-DRL is able to 1) tolerate unknown-unknowns disturbances, 2) guarantee mathematically provable safety and stability, and 3) strictly comply with physical knowledge pertaining to Bellman equation and reward. The effectiveness of the Phy-DRL is finally validated by an inverted pendulum and a quadruped robot. The experimental results demonstrate that compared with purely data-driven DRL, Phy-DRL features remarkably fewer learning parameters, accelerated training and enlarged reward, while offering enhanced model robustness and safety assurance.

It is commonly perceived that the strongest language models (LMs) rely on a combination of massive scale, instruction data, and human feedback to perform specialized tasks -- e.g. summarization and paraphrasing, without supervision. In this paper, we propose that language models can learn to summarize and paraphrase sentences, with none of these 3 factors. We present Impossible Distillation, a framework that distills a task-specific dataset directly from an off-the-shelf LM, even when it is impossible for the LM itself to reliably solve the task. By training a student model on the generated dataset and amplifying its capability through self-distillation, our method yields a high-quality model and dataset from a low-quality teacher model, without the need for scale or supervision. Using Impossible Distillation, we are able to distill an order of magnitude smaller model (with only 770M parameters) that outperforms 175B parameter GPT-3, in both quality and controllability, as confirmed by automatic and human evaluations. Furthermore, as a useful byproduct of our approach, we obtain DIMSUM+, a high-quality dataset with 3.4M sentence summaries and paraphrases. Our analyses show that this dataset, as a purely LM-generated corpus, is more diverse and more effective for generalization to unseen domains than all human-authored datasets -- including Gigaword with 4M samples.

Abstract--For many robotic manipulation and contact tasks, it is crucial to accurately estimate uncertain object poses, for which certain geometry and sensor information are fused in some optimal fashion. Previous results for this problem primarily adopt sampling-based or end-to-end learning methods, which yet often suffer from the issues of efficiency and generalizability. In this paper, we propose a novel differentiable framework for this uncertain pose estimation during contact, so that it can be solved in an efficient and accurate manner with gradient-based solver. To achieve this, we introduce a new geometric definition that is highly adaptable and capable of providing differentiable contact Figure 1: Graphical abstracts illustrating our differentiable pose estimation features. Then we approach the problem from a bi-level perspective during contact. Left: A peg-in-hole task performed in a hole with and utilize the gradient of these contact features along with pose uncertainty along the x and y directions. Right: Visualization of differentiable optimization to efficiently solve for the uncertain the differentiable cost landscape and the gradient-based optimization pose. Several scenarios are implemented to demonstrate how the process utilizing force/torque sensor information acquired through proposed framework can improve existing methods.

Modal parameter estimation of operational structures is often a challenging task when confronted with unwanted distortions (outliers) in field measurements. Atypical observations present a problem to operational modal analysis (OMA) algorithms, such as stochastic subspace identification (SSI), severely biasing parameter estimates and resulting in misidentification of the system. Despite this predicament, no simple mechanism currently exists capable of dealing with such anomalies in SSI. Addressing this problem, this paper first introduces a novel probabilistic formulation of stochastic subspace identification (Prob-SSI), realised using probabilistic projections. Mathematically, the equivalence between this model and the classic algorithm is demonstrated. This fresh perspective, viewing SSI as a problem in probabilistic inference, lays the necessary mathematical foundation to enable a plethora of new, more sophisticated OMA approaches. To this end, a statistically robust SSI algorithm (robust Prob-SSI) is developed, capable of providing a principled and automatic way of handling outlying or anomalous data in the measured timeseries, such as may occur in field recordings, e.g. intermittent sensor dropout. Robust Prob-SSI is shown to outperform conventional SSI when confronted with 'corrupted' data, exhibiting improved identification performance and higher levels of confidence in the found poles when viewing consistency (stabilisation) diagrams. Similar benefits are also demonstrated on the Z24 Bridge benchmark dataset, highlighting enhanced performance on measured systems.

Model-free reinforcement learning algorithms have exhibited great potential in solving single-task sequential decision-making problems with high-dimensional observations and long horizons, but are known to be hard to generalize across tasks. Model-based RL, on the other hand, learns task-agnostic models of the world that naturally enables transfer across different reward functions, but struggles to scale to complex environments due to the compounding error. To get the best of both worlds, we propose a self-supervised reinforcement learning method that enables the transfer of behaviors across tasks with different rewards, while circumventing the challenges of model-based RL. In particular, we show self-supervised pre-training of model-free reinforcement learning with a number of random features as rewards allows implicit modeling of long-horizon environment dynamics. Then, planning techniques like model-predictive control using these implicit models enable fast adaptation to problems with new reward functions. Our method is self-supervised in that it can be trained on offline datasets without reward labels, but can then be quickly deployed on new tasks. We validate that our proposed method enables transfer across tasks on a variety of manipulation and locomotion domains in simulation, opening the door to generalist decision-making agents.

This paper develops computationally efficient data-driven model predictive control (MPC) for Agile quadrotor flight. Agile quadrotors in high-speed flights can experience high levels of aerodynamic effects. Modeling these turbulent aerodynamic effects is a cumbersome task and the resulting model may be overly complex and computationally infeasible. Combining Gaussian Process (GP) regression models with a simple dynamic model of the system has demonstrated significant improvements in control performance. However, direct integration of the GP models to the MPC pipeline poses a significant computational burden to the optimization process. Therefore, we present an approach to separate the GP models to the MPC pipeline by computing the model corrections using reference trajectory and the current state measurements prior to the online MPC optimization. This method has been validated in the Gazebo simulation environment and has demonstrated of up to $50\%$ reduction in trajectory tracking error, matching the performance of the direct GP integration method with improved computational efficiency.