Large Language Models (LLMs) have achieved remarkable results. However, existing models are expensive to train and deploy, and it is also difficult to expand their knowledge beyond pre-training data without forgetting previous knowledge. This paper proposes a new neural network architecture, ModuleFormer, that leverages modularity to improve the efficiency and flexibility of large language models. ModuleFormer is based on the Sparse Mixture of Experts (SMoE). Unlike the previous SMoE-based modular language model, which requires domain-labeled data to learn domain-specific experts, ModuleFormer can induce modularity from uncurated data with its new load balancing and concentration losses. ModuleFormer is a modular architecture that includes two different types of modules: new stick-breaking attention heads and feedforward experts. Different modules are sparsely activated conditions on the input token during training and inference. In our experiment, we found that the modular architecture enables three important abilities for large pre-trained language models: 1) Efficiency, since ModuleFormer only activates a subset of its modules for each input token, thus it could achieve the same performance as dense LLMs with more than two times throughput; 2) Extendability, ModuleFormer is more immune to catastrophic forgetting than dense LLMs and can be easily extended with new modules to learn new knowledge that is not included in the training data; 3) Specialisation, finetuning ModuleFormer could specialize a subset of modules to the finetuning task and the task-unrelated modules could be easily pruned for a lightweight deployment.

Inexpensive machine learning potentials are increasingly being used to speed up structural optimization and molecular dynamics simulations of materials by iteratively predicting and applying interatomic forces. In these settings, it is crucial to detect when predictions are unreliable to avoid wrong or misleading results. Here, we present a complete framework for training and recalibrating graph neural network ensemble models to produce accurate predictions of energy and forces with calibrated uncertainty estimates. The proposed method considers both epistemic and aleatoric uncertainty and the total uncertainties are recalibrated post hoc using a nonlinear scaling function to achieve good calibration on previously unseen data, without loss of predictive accuracy. The method is demonstrated and evaluated on two challenging, publicly available datasets, ANI-1x (Smith et al.) and Transition1x (Schreiner et al.), both containing diverse conformations far from equilibrium. A detailed analysis of the predictive performance and uncertainty calibration is provided. In all experiments, the proposed method achieved low prediction error and good uncertainty calibration, with predicted uncertainty correlating with expected error, on energy and forces. To the best of our knowledge, the method presented in this paper is the first to consider a complete framework for obtaining calibrated epistemic and aleatoric uncertainty predictions on both energy and forces in ML potentials.

For Prognostics and Health Management (PHM) of Lithium-ion (Li-ion) batteries, many models have been established to characterize their degradation process. The existing empirical or physical models can reveal important information regarding the degradation dynamics. However, there are no general and flexible methods to fuse the information represented by those models. Physics-Informed Neural Network (PINN) is an efficient tool to fuse empirical or physical dynamic models with data-driven models. To take full advantage of various information sources, we propose a model fusion scheme based on PINN. It is implemented by developing a semi-empirical semi-physical Partial Differential Equation (PDE) to model the degradation dynamics of Li-ion batteries. When there is little prior knowledge about the dynamics, we leverage the data-driven Deep Hidden Physics Model (DeepHPM) to discover the underlying governing dynamic models. The uncovered dynamics information is then fused with that mined by the surrogate neural network in the PINN framework. Moreover, an uncertainty-based adaptive weighting method is employed to balance the multiple learning tasks when training the PINN. The proposed methods are verified on a public dataset of Li-ion Phosphate (LFP)/graphite batteries.

Model-based reinforcement learning (MBRL) with real-time planning has shown great potential in locomotion and manipulation control tasks. However, the existing planning methods, such as the Cross-Entropy Method (CEM), do not scale well to complex high-dimensional environments. One of the key reasons for underperformance is the lack of exploration, as these planning methods only aim to maximize the cumulative extrinsic reward over the planning horizon. Furthermore, planning inside the compact latent space in the absence of observations makes it challenging to use curiosity-based intrinsic motivation. We propose Curiosity CEM (CCEM), an improved version of the CEM algorithm for encouraging exploration via curiosity. Our proposed method maximizes the sum of state-action Q values over the planning horizon, in which these Q values estimate the future extrinsic and intrinsic reward, hence encouraging reaching novel observations. In addition, our model uses contrastive representation learning to efficiently learn latent representations. Experiments on image-based continuous control tasks from the DeepMind Control suite show that CCEM is by a large margin more sample-efficient than previous MBRL algorithms and compares favorably with the best model-free RL methods.

Ensuring safety and meeting temporal specifications are critical challenges for long-term robotic tasks. Signal temporal logic (STL) has been widely used to systematically and rigorously specify these requirements. However, traditional methods of finding the control policy under those STL requirements are computationally complex and not scalable to high-dimensional or systems with complex nonlinear dynamics. Reinforcement learning (RL) methods can learn the policy to satisfy the STL specifications via hand-crafted or STL-inspired rewards, but might encounter unexpected behaviors due to ambiguity and sparsity in the reward. In this paper, we propose a method to directly learn a neural network controller to satisfy the requirements specified in STL. Our controller learns to roll out trajectories to maximize the STL robustness score in training. In testing, similar to Model Predictive Control (MPC), the learned controller predicts a trajectory within a planning horizon to ensure the satisfaction of the STL requirement in deployment. A backup policy is designed to ensure safety when our controller fails. Our approach can adapt to various initial conditions and environmental parameters. We conduct experiments on six tasks, where our method with the backup policy outperforms the classical methods (MPC, STL-solver), model-free and model-based RL methods in STL satisfaction rate, especially on tasks with complex STL specifications while being 10X-100X faster than the classical methods.

Line outage identification in distribution grids is essential for sustainable grid operation. In this work, we propose a practical yet robust detection approach that utilizes only readily available voltage magnitudes, eliminating the need for costly phase angles or power flow data. Given the sensor data, many existing detection methods based on change-point detection require prior knowledge of outage patterns, which are unknown for real-world outage scenarios. To remove this impractical requirement, we propose a data-driven method to learn the parameters of the post-outage distribution through gradient descent. However, directly using gradient descent presents feasibility issues. To address this, we modify our approach by adding a Bregman divergence constraint to control the trajectory of the parameter updates, which eliminates the feasibility problems. As timely operation is the key nowadays, we prove that the optimal parameters can be learned with convergence guarantees via leveraging the statistical and physical properties of voltage data. We evaluate our approach using many representative distribution grids and real load profiles with 17 outage configurations. The results show that we can detect and localize the outage in a timely manner with only voltage magnitudes and without assuming a prior knowledge of outage patterns.

Drones are becoming versatile in a myriad of applications. This has led to the use of drones for spying and intruding into the restricted or private air spaces. Such foul use of drone technology is dangerous for the safety and security of many critical infrastructures. In addition, due to the varied low-cost design and agility of the drones, it is a challenging task to identify and track them using the conventional radar systems. In this paper, we propose a reinforcement learning based approach for identifying and tracking any intruder drone using a chaser drone. Our proposed solution uses computer vision techniques interleaved with the policy learning framework of reinforcement learning to learn a control policy for chasing the intruder drone. The whole system has been implemented using ROS and Gazebo along with the Ardupilot based flight controller. The results show that the reinforcement learning based policy converges to identify and track the intruder drone. Further, the learnt policy is robust with respect to the change in speed or orientation of the intruder drone.

As satellites become smaller, the ability to maintain stable pointing decreases as external forces acting on the satellite come into play. At the same time, reaction wheels used in the attitude determination and control system (ADCS) introduce high frequency jitter which can disrupt pointing stability. For space domain awareness (SDA) tasks that track objects tens of thousands of kilometres away, the pointing accuracy offered by current nanosats, typically in the range of 10 to 100 arcseconds, is not sufficient. In this work, we develop a novel payload that utilises a neuromorphic event sensor (for high frequency and highly accurate relative attitude estimation) paired in a closed loop with a piezoelectric stage (for active attitude corrections) to provide highly stable sensor-specific pointing. Event sensors are especially suited for space applications due to their desirable characteristics of low power consumption, asynchronous operation, and high dynamic range. We use the event sensor to first estimate a reference background star field from which instantaneous relative attitude is estimated at high frequency. The piezoelectric stage works in a closed control loop with the event sensor to perform attitude corrections based on the discrepancy between the current and desired attitude. Results in a controlled setting show that we can achieve a pointing accuracy in the range of 1-5 arcseconds using our novel payload at an operating frequency of up to 50Hz using a prototype built from commercial-off-the-shelf components. Further details can be found at https://ylatif.github.io/ultrafinestabilisation

This paper presents an analysis of parametric characterization of a motor driven tendon-sheath actuator system for use in upper limb augmentation for applications such as rehabilitation, therapy, and industrial automation. The double tendon sheath system, which uses two sets of cables (agonist and antagonist side) guided through a sheath, is considered to produce smooth and natural-looking movements of the arm. The exoskeleton is equipped with a single motor capable of controlling both the flexion and extension motions. One of the key challenges in the implementation of a double tendon sheath system is the possibility of slack in the tendon, which can impact the overall performance of the system. To address this issue, a robust mathematical model is developed and a comprehensive parametric study is carried out to determine the most effective strategies for overcoming the problem of slack and improving the transmission. The study suggests that incorporating a series spring into the system's tendon leads to a universally applicable design, eliminating the need for individual customization. The results also show that the slack in the tendon can be effectively controlled by changing the pretension, spring constant, and size and geometry of spool mounted on the axle of motor.

In this paper, we discuss the Dutch power market, which is comprised of a day-ahead market and an intraday balancing market that operates like an auction. Due to fluctuations in power supply and demand, there is often an imbalance that leads to different prices in the two markets, providing an opportunity for arbitrage. To address this issue, we restructure the problem and propose a collaborative dual-agent reinforcement learning approach for this bi-level simulation and optimization of European power arbitrage trading. We also introduce two new implementations designed to incorporate domain-specific knowledge by imitating the trading behaviours of power traders. By utilizing reward engineering to imitate domain expertise, we are able to reform the reward system for the RL agent, which improves convergence during training and enhances overall performance. Additionally, the tranching of orders increases bidding success rates and significantly boosts profit and loss (P&L). Our study demonstrates that by leveraging domain expertise in a general learning problem, the performance can be improved substantially, and the final integrated approach leads to a three-fold improvement in cumulative P&L compared to the original agent. Furthermore, our methodology outperforms the highest benchmark policy by around 50% while maintaining efficient computational performance.