Search for: All records

Abstract This paper considers a robot moving in a 3D environment that is tasked with estimating a quasi-stationary environmental field (e.g., temperature, concentration of a chemical pollutant, or distribution of light radiation density) in the presence of localization uncertainties, as is typical in underwater or other GPS-denied environments. Gaussian process regression has been widely adopted to model environmental fields. However, a drawback of Gaussian process regression is its difficulty in accounting for data with uncertain input. This work proposes a novel multi-fidelity Gaussian process-based regression approach to address the challenge by splitting the data collected by the robot into different datasets corresponding to the amount of input (localization) uncertainty. Furthermore, a sampling-based trajectory planning algorithm is proposed for adaptive robot exploration that optimizes a field-reconstruction objective function while accommodating resource constraints. The proposed approach is experimentally evaluated using a miniature gliding robotic fish that measures light intensity in a large indoor tank. The adaptive exploration algorithm is tested using both a multi-fidelity Gaussian process model and a baseline single-fidelity model. Two objective functions, based on the information gain and an ergodic metric, respectively, are adopted in the evaluation. The experiments show that, for both objective functions, using multi-fidelity Gaussian process reduces the weighted mean squared error between the model prediction and the ground-truth field compared to using the baseline single-fidelity model that ignores localization uncertainty. Accompanying code available at Coleman (Adaptive exploration under localization uncertainty using multi-fidelity Gaussian processes, 2025,https://github.com/colem404/Adaptive-Exploration-Under-Localization-Uncertainty-Using-Multi-fidelity-Gaussian-Processes/tree/main). 

Abstract Drifters are energy-efficient sampling platforms widely used in rivers, lakes, and oceans with pronounced flows. Passive drifters cannot exert control over their motion, so their motion is determined by the local flow structure. To overcome this limitation, we propose an active drifter with a thruster to generate propulsion and increase maneuverability. To retain the energy-efficient nature of drifters, the application of the thruster must be carefully planned. In this paper, we consider an optimal control problem for an active drifter with one degree-of-freedom, where its motion is governed by rudder-modulated drag and thruster propulsion. Despite the realistic and complex nonlinear dynamics, an analytical solution is developed using Pontryagin's minimum principle (PMP). Optimal solutions are back-propagated from the final state to find state-space locations where the optimal control switches. Such locations are parameterized by one of the costate variables and are related to the system's final state. These locations enclose regions in the state space where the optimal control is the same; thus, the optimal control can be mapped to the state space to be used as a feedback control law that depends on the final state. The proposed approach is first illustrated in simulation, and the tradeoff between thruster use and time taken is examined. The efficacy of the proposed approach is further demonstrated by results from field experiments conducted with an active drifter prototype. 

Mobile sampling platforms are increasingly being used in marine studies. Accurate control of such platforms is often essential to study design, but is also difficult as the dynamics of mobile platforms are often nonlinear, nonholonomic, and underactuated. Event-triggered control where the applied control is updated, often aperiodically, at discrete times, addresses practical limitations on computational power and actuator response; however, most of the existing results ignore nonholonomic constraints and assume input-to-state stability of the closed-loop system. This work proposes a novel event-trigger design that relaxes these limitations. The efficacy of the proposed event-triggering mechanism is further shown in simulation. 

Soft robots, distinguished by their inherent compliance and continuum structures, present unique modeling challenges, especially when subjected to significant external loads such as gravity and payloads. In this study, we introduce an innovative data-driven modeling framework leveraging an Euler spiral-inspired shape representations to accurately describe the complex shapes of soft continuum actuators. Based on this representation, we develop neural network-based forward and inverse models to effectively capture the nonlinear behavior of a fiber-reinforced pneumatic bending actuator. Our forward model accurately predicts the actuator’s deformation given inputs of pressure and payload, while the inverse model reliably estimates payloads from observed actuator shapes and known pressure inputs. Comprehensive experimental validation demonstrates the effectiveness and accuracy of our proposed approach. Notably, the augmented Euler spiral-based forward model achieves low average positional prediction errors of 3.38%, 2.19%, and 1.93% of the actuator length at the one-third, two-thirds, and tip positions, respectively. Furthermore, the inverse model demonstrates precision of estimating payloads with an average error as low as 0.72% across the tested range. These results underscore the potential of our method to significantly enhance the accuracy and predictive capabilities of modeling frameworks for soft robotic systems. 

Data-driven model predictive control (MPC) has demonstrated significant potential for improving robot control performance in the presence of model uncertainties. However, existing approaches often require extensive offline data collection and computationally intensive training, limiting their ability to adapt online. To address these challenges, this paper presents a fast online adaptive MPC framework that leverages neural networks integrated with Model-Agnostic Meta-Learning (MAML). Our approach focuses on few-shot adaptation of residual dynamics—capturing the discrepancy between nominal and true system behavior—using minimal online data and gradient steps. By embedding these meta-learned residual models into a computationally efficient L4CasADi-based MPC pipeline, the proposed method enables rapid model correction, enhances predictive accuracy, and improves real-time control performance. We validate the framework through simulation studies on a Cart-Pole system and a 2D quadrotor. Results show significant gains in adaptation speed and prediction accuracy over both nominal MPC and nominal MPC augmented with a freshly initialized neural network, underscoring the effectiveness of our approach for real-time adaptive robot control. 

Abstract Stretchable pressure and strain sensors have been the focus of extensive research in recent years due to their significant potential. Several studies have explored different approaches to enhance their performance. However, most of these works focus on separate sensors for each purpose, and the research investigating the simultaneous measurement of both pressure and strain is limited. This paper presents a stretchable capacitive sensor capable of simultaneously measuring pressure (up to 245.5 kPa) and strain (up to 40%). Our design leverages the conductivity and piezoresistivity of an inexpensive carbon black and Ecoflex 00-30 composite, which serves as both the strain-detection element and the electrodes for the capacitive pressure detection element. To characterize the sensor’s response to both stimuli, we apply a range of pressures and strains concurrently and observed the corresponding capacitance and resistance changes. The characterization results show that the measured resistance change is primarily determined by the applied strain, while the measured capacitance change is dependent on both the pressure and the strain. A computationally efficient polynomial model is identified to capture the capacitance change as a joint function of the pressure and strain inputs, which enables the estimation of the pressure once the strain is inferred from the resistance measurement. Validation experiments show that the approach to simultaneous measurement of strain and pressure is feasible and effective. The proposed work holds promise for advances in soft robotics, wearable electronics, and biomedical monitoring. 

While recent advancements in motor learning have emphasized the critical role of systematic task scheduling in enhancing task learning, the heuristic design of task schedules remains predominant. Random task scheduling can lead to sub-optimal motor learning, whereas performance-based scheduling might not be adequate for complex motor skill acquisition. This paper addresses these challenges by proposing a model-based approach for online skill estimation and individualized task scheduling in de-novo (novel) motor learning tasks. We introduce a framework utilizing a personalized human motor learning model and particle filter for skill state estimation, coupled with a stochastic nonlinear model predictive control (SNMPC) strategy to optimize curriculum design for a high-dimensional motor task. Simulation results show the effectiveness of our framework in estimating the latent skill state, and the efficacy of the framework in accelerating skill learning. Furthermore, a human subject study shows that the group with the SNMPC-based curriculum design exhibited expedited skill learning and improved task performance. Our contributions offer a pathway towards expedited motor learning across various novel tasks, with implications for enhancing rehabilitation and skill acquisition processes. 

Abstract With advances in materials and manufacturing techniques, recent years have seen a number of conductive composite materials that exhibit pronounced strain-dependent electrical resistivity, allowing them to be used for embedded, cost-effective strain sensing in various applications. The strain-resistivity relationship of these materials, however, is often highly nonlinear and dynamic, posing challenges for effective use of such strain sensors. In this paper, a computationally efficient scheme is proposed for compensating the nonlinear, dynamic strain-resistance behavior of a soft conductive rubber using a time delay neural network. The accuracy and feasibility of the technique is evaluated with a soft robotic arm incorporating three strain sensors for proprioception. Experimental results show that the sensing scheme is able to predict both the tip position and the shape of the robotic manipulator, achieving an average tip positional error of less than 4% relative to the total length of the manipulator.