Manipulating matter with a scanning tunneling microscope (STM) enables creation of atomically defined artificial structures that host designer quantum states. However, the time-consuming nature of the manipulation process, coupled with the sensitivity of the STM tip, constrains the exploration of diverse configurations and limits the size of designed features. In this study, we present a reinforcement learning (RL)-based framework for creating artificial structures by spatially manipulating carbon monoxide (CO) molecules on a copper substrate using the STM tip. The automated workflow combines molecule detection and manipulation, employing deep learning-based object detection to locate CO molecules and linear assignment algorithms to allocate these molecules to designated target sites. We initially perform molecule maneuvering based on randomized parameter sampling for sample bias, tunneling current setpoint and manipulation speed. This dataset is then structured into an action trajectory used to train an RL agent. The model is subsequently deployed on the STM for real-time fine-tuning of manipulation parameters during structure construction. Our approach incorporates path planning protocols coupled with active drift compensation to enable atomically precise fabrication of structures with significantly reduced human input while realizing larger-scale artificial lattices with desired electronic properties. Using our approach, we demonstrate the automated construction of an extended artificial graphene lattice and confirm the existence of characteristic Dirac point in its electronic structure. Further challenges to RL-based structural assembly scalability are discussed.

Discovering symbolic representations for skills is essential for abstract reasoning and efficient planning in robotics. Previous neuro-symbolic robotic studies mostly focused on discovering perceptual symbolic categories given a pre-defined action repertoire and generating plans with given action symbols. A truly developmental robotic system, on the other hand, should be able to discover all the abstractions required for the planning system with minimal human intervention. In this study, we propose a novel system that is designed to discover symbolic action primitives along with perceptual symbols autonomously. Our system is based on an encoder-decoder structure that takes object and action information as input and predicts the generated effect. To efficiently explore the vast continuous action parameter space, we introduce a Curiosity-Based exploration module that selects the most informative actions -- the ones that maximize the entropy in the predicted effect distribution. The discovered symbolic action primitives are then used to make plans using a symbolic tree search strategy in single- and double-object manipulation tasks. We compare our model with two baselines that use different exploration strategies in different experiments. The results show that our approach can learn a diverse set of symbolic action primitives, which are effective for generating plans in order to achieve given manipulation goals.

Adaptive traffic signal control (ATSC) can effectively reduce vehicle travel times by dynamically adjusting signal timings but poses a critical challenge in real-world scenarios due to the complexity of real-time decision-making in dynamic and uncertain traffic conditions. The burgeoning field of intelligent transportation systems, bolstered by artificial intelligence techniques and extensive data availability, offers new prospects for the implementation of ATSC. In this study, we introduce a parallel hybrid action space reinforcement learning model (PH-DDPG) that optimizes traffic signal phase and duration of traffic signals simultaneously, eliminating the need for sequential decision-making seen in traditional two-stage models. Our model features a task-specific parallel hybrid action space tailored for adaptive traffic control, which directly outputs discrete phase selections and their associated continuous duration parameters concurrently, thereby inherently addressing dynamic traffic adaptation through unified parametric optimization. %Our model features a unique parallel hybrid action space that allows for the simultaneous output of each action and its optimal parameters, streamlining the decision-making process. Furthermore, to ascertain the robustness and effectiveness of this approach, we executed ablation studies focusing on the utilization of a random action parameter mask within the critic network, which decouples the parameter space for individual actions, facilitating the use of preferable parameters for each action. The results from these studies confirm the efficacy of this method, distinctly enhancing real-world applicability

Dynamic in-hand manipulation remains a challenging task for soft robotic systems that have demonstrated advantages in safe compliant interactions but struggle with high-speed dynamic tasks. In this work, we present SWIFT, a system for learning dynamic tasks using a soft and compliant robotic hand. Unlike previous works that rely on simulation, quasi-static actions and precise object models, the proposed system learns to spin a pen through trial-and-error using only real-world data without requiring explicit prior knowledge of the pen's physical attributes. With self-labeled trials sampled from the real world, the system discovers the set of pen grasping and spinning primitive parameters that enables a soft hand to spin a pen robustly and reliably. After 130 sampled actions per object, SWIFT achieves 100% success rate across three pens with different weights and weight distributions, demonstrating the system's generalizability and robustness to changes in object properties. The results highlight the potential for soft robotic end-effectors to perform dynamic tasks including rapid in-hand manipulation. We also demonstrate that SWIFT generalizes to spinning items with different shapes and weights such as a brush and a screwdriver which we spin with 10/10 and 5/10 success rates respectively. Videos, data, and code are available at https://soft-spin.github.io.

Large Language Models (LLMs) present a promising frontier in robotic task planning by leveraging extensive human knowledge. Nevertheless, the current literature often overlooks the critical aspects of adaptability and error correction within robotic systems. This work aims to overcome this limitation by enabling robots to modify their motion strategies and select the most suitable task plans based on the context. We introduce a novel framework termed action contextualization, aimed at tailoring robot actions to the precise requirements of specific tasks, thereby enhancing adaptability through applying LLM-derived contextual insights. Our proposed motion metrics guarantee the feasibility and efficiency of adjusted motions, which evaluate robot performance and eliminate planning redundancies. Moreover, our framework supports online feedback between the robot and the LLM, enabling immediate modifications to the task plans and corrections of errors. Our framework has achieved an overall success rate of 81.25% through extensive validation. Finally, integrated with dynamic system (DS)-based robot controllers, the robotic arm-hand system demonstrates its proficiency in autonomously executing LLM-generated motion plans for sequential table-clearing tasks, rectifying errors without human intervention, and completing tasks, showcasing robustness against external disturbances. Our proposed framework features the potential to be integrated with modular control approaches, significantly enhancing robots' adaptability and autonomy in sequential task execution.

Conventional Task and Motion Planning (TAMP) approaches rely on manually crafted interfaces connecting symbolic task planning with continuous motion generation. These domain-specific and labor-intensive modules are limited in addressing emerging tasks in real-world settings. Here, we present LLM^3, a novel Large Language Model (LLM)-based TAMP framework featuring a domain-independent interface. Specifically, we leverage the powerful reasoning and planning capabilities of pre-trained LLMs to propose symbolic action sequences and select continuous action parameters for motion planning. Crucially, LLM^3 incorporates motion planning feedback through prompting, allowing the LLM to iteratively refine its proposals by reasoning about motion failure. Consequently, LLM^3 interfaces between task planning and motion planning, alleviating the intricate design process of handling domain-specific messages between them. Through a series of simulations in a box-packing domain, we quantitatively demonstrate the effectiveness of LLM^3 in solving TAMP problems and the efficiency in selecting action parameters. Ablation studies underscore the significant contribution of motion failure reasoning to the success of LLM^3. Furthermore, we conduct qualitative experiments on a physical manipulator, demonstrating the practical applicability of our approach in real-world settings.

Optimizing robotic action parameters is a significant challenge for manipulation tasks that demand high levels of precision and generalization. Using a model-based approach, the robot must quickly reason about the outcomes of different actions using a predictive model to find a set of parameters that will have the desired effect. The model may need to capture the behaviors of rigid and deformable objects, as well as objects of various shapes and sizes. Predictive models often need to trade-off speed for prediction accuracy and generalization. This paper proposes a framework that leverages the strengths of multiple predictive models, including analytical, learned, and simulation-based models, to enhance the efficiency and accuracy of action parameter optimization. Our approach uses Model Deviation Estimators (MDEs) to determine the most suitable predictive model for any given state-action parameters, allowing the robot to select models to make fast and precise predictions. We extend the MDE framework by not only learning sim-to-real MDEs, but also sim-to-sim MDEs. Our experiments show that these sim-to-sim MDEs provide significantly faster parameter optimization as well as a basis for efficiently learning sim-to-real MDEs through finetuning. The ease of collecting sim-to-sim training data also allows the robot to learn MDEs based directly on visual inputs and local material properties.

In haptic object discrimination, the effect of gripper embodiment, action parameters, and sensory channels has not been systematically studied. We used two anthropomorphic hands and two 2-finger grippers to grasp two sets of deformable objects. On the object classification task, we found: (i) among classifiers, SVM on sensory features and LSTM on raw time series performed best across all grippers; (ii) faster compression speeds degraded performance; (iii) generalization to different grasping configurations was limited; transfer to different compression speeds worked well for the Barrett Hand only. Visualization of the feature spaces using PCA showed that the gripper morphology and the action parameters were the main source of variance, rendering generalization across embodiment or grasp configurations very hard. On the highly challenging dataset consisting of polyurethane foams alone, only the Barrett Hand achieved excellent performance. Tactile sensors can thus provide a key advantage even if recognition is based on stiffness rather than shape. The dataset with 24000 measurements is publicly available.

This paper presents a method using data-driven models for selecting actions and predicting the total performance of autonomous wheel loader operations over many loading cycles in a changing environment. The performance includes loaded mass, loading time, work. The data-driven models input the control parameters of a loading action and the heightmap of the initial pile state to output the inference of either the performance or the resulting pile state. By iteratively utilizing the resulting pile state as the initial pile state for consecutive predictions, the prediction method enables long-horizon forecasting. Deep neural networks were trained on data from over 10,000 random loading actions in gravel piles of different shapes using 3D multibody dynamics simulation. The models predict the performance and the resulting pile state with, on average, 95% accuracy in 1.2 ms, and 97% in 4.5 ms, respectively. The performance prediction was found to be even faster in exchange for accuracy by reducing the model size with the lower dimensional representation of the pile state using its slope and curvature. The feasibility of long-horizon predictions was confirmed with 40 sequential loading actions at a large pile. With the aid of a physics-based model, the pile state predictions are kept sufficiently accurate for longer-horizon use.

A contextual bandit is a popular framework for online learning to act under uncertainty. In practice, the number of actions is huge and their expected rewards are correlated. In this work, we introduce a general framework for capturing such correlations through a mixed-effect model where actions are related through multiple shared effect parameters. To explore efficiently using this structure, we propose Mixed-Effect Thompson Sampling (meTS) and bound its Bayes regret. The regret bound has two terms, one for learning the action parameters and the other for learning the shared effect parameters. The terms reflect the structure of our model and the quality of priors. Our theoretical findings are validated empirically using both synthetic and real-world problems. We also propose numerous extensions of practical interest. While they do not come with guarantees, they perform well empirically and show the generality of the proposed framework.