Integrating deep learning with environmental perception enhances robotic adaptability to complex tasks. However, its “black-box” nature, such as the lack of uncertainty quantification, poses challenges for safety-critical applications, particularly in unstructured and noisy environments. Bayesian neural networks (BNNs) offer uncertainty quantification but are limited by high hardware overhead, restricting real-time implementation on resource-constrained robots. This paper presents a mixedsignal hardware accelerator for BNNs, utilizing probabilistic quantum tunneling in fully depleted silicon-on-insulator (FDSOI) transistors to enable efficient, real-time uncertainty quantification. Device measurements indicate high-quality Gaussian random variable generation, validated through quantile-quantile plot analysis, with a high correlation coefficient ($r=0.997$) at $200 \mathrm{fJ} /$ sample. Leveraging such compact randomness, the parallel architecture achieved $10^{3}-10^{4} \times$ latency reduction at less than $2 \times$ area cost. Finally, in uncertainty-aware visual localization application of autonomous underwater vehicles, the BNN model effectively distinguishes data noise from model uncertainty, yielding significant information gain and enhancing the resampling efficiency by $4.5 \times$ at same accuracy.

Underwater robotic manipulation faces significant challenges due to complex fluid dynamics and unstructured environments, causing most manipulation systems to rely heavily on human teleoperation. In this paper, we introduce AquaBot, a fully autonomous manipulation system that combines behavior cloning from human demonstrations with self-learning optimization to improve beyond human teleoperation performance. With extensive real-world experiments, we demonstrate AquaBot's versatility across diverse manipulation tasks, including object grasping, trash sorting, and rescue retrieval. Our real-world experiments show that AquaBot's self-optimized policy outperforms a human operator by 41% in speed. AquaBot represents a promising step towards autonomous and self-improving underwater manipulation systems. We will open-source both hardware and software implementation details.

For a class of unknown MIMO non-affine nonlinear repetitive discrete-time systems, a novel data-driven high-order point-to-point iterative learning control scheme is proposed. The control input objective function of this method consists of two parts. One includes the high-order error information, the other consists of the control inputs within the time sub-intervals divided by prescribed desired points. The control law is designed by optimizing this function and it comprises only the known control input signals in the current iteration and the error data in previous iterations. Further, the convergence analysis is conducted in a data-driven way and does not need precise mathematical models. In addition, a scalar index function is set up to evaluate the tracking error convergence rate. By choosing the appropriate high-order factor and corresponding step-size factors, the convergence rate of higher-order learning law is shown to have a faster speed than that of lower-order one. Simulation experiments verify the effectiveness and advantage of this method.Note to Practitioners—The motivation of this paper is to design a control algorithm that only depends on the information of the prescribed desired points, that is, the point-to-point iterative learning control scheme. This control algorithm can be competent for terminal temperature control tasks with very intense chemical reactions and other control tasks that the information except for the I/O data at the prescribed desired points is unavailable. When we want to improve further the control performance at some prescribed desired points, the data of some intermediate time instants can be used to supply extra support for the control system design to improve the control performance at the prescribed desired points. In fact, the point-to-point iterative learning control scheme can be applied to many fields, such as high-speed trains, functional electrical stimulation areas, and positioning X–Y tables. In this paper, a data-driven high-order point-to-point iterative learning control scheme for the MIMO systems is designed, and the stability of this scheme is theoretically analyzed by using the contraction mapping method. Further, the tracking error convergence rate with different order learning law is analyzed. Finally, two numerical simulations are used to verify the effectiveness of the scheme proposed in this paper.

In this article, a disturbance observer dynamic linearization (DL)-based model-free adaptive control (MFAC) scheme is proposed for discrete-time nonlinear systems with disturbances and uncertainties. The partial-form-dynamic-linearization-based disturbance observer (PDO) is constructed by applying the DL method to an unknown ideal disturbance observer. An adaptive updating algorithm of the observer gain is derived by minimizing a estimation criterion function. Then, the PDO-based MFAC scheme is formed and its bounded stability is rigorously analyzed using the contraction mapping principle. The proposed scheme is a purely data-driven control method, that is, both the PDO and control system are designed by using only the input/output data of underlying system. A numerical simulation and a vehicle turning experiment are given to verify the effectiveness of the proposed scheme.

We propose a memory-constrained partition-based method to extract symbolic representations of the belief state and its dynamics in order to solve planning problems in a partially observable Markov decision process (POMDP). Our K-means partitioning strategy uses a fixed number of symbols to represent the partitions of the belief space and ensures the parameterization of the belief dynamics does not grow exponentially as the system dimension increases. By casting our problem as a partitioning of the POMDP, we can then solve planning problems using traditional symbolic planning solvers (such as HTN or A* solvers). Our work is motivated by an autonomous underwater vehicle navigation problem where the vehicle is affected by uncertain flow conditions and receives severely limited position observations. Simulation experiments are provided to validate the performance of the proposed algorithms.

Swing oscillation is widely observed among indoor miniature autonomous blimps (MABs) due to their underactuated design and unique aerodynamic shape. A detailed dynamics model is critical for investigating this undesired movement and designing controllers to stabilize the oscillation. This paper presents a motion model that describes the coupled translational and rotational movements of a typical indoor MAB with saucer-shaped envelope. The kinematics and dynamic model of the MAB are simplified from the six-degrees-of-freedom (6-DOF) Newton-Euler equations of underwater vehicles. The model is then reduced to 3-DOF given the symmetrical design of the MAB around its vertical axis. Parameters of the motion model are estimated from the system identification experiments, and validated with experimental data.

Indoor miniature autonomous blimp (MAB) is a small-sized aerial platform with outstanding safety and flight endurance. A detailed six-degree-of-freedom (6DOF) dynamics model is critical for controller design and motion simulation. This paper presents the identification of the rotation-related parameters of the blimp dynamics model through swing motion of the robot. A pendulum-like grey box model is constructed to identify the parameters from physical measurements and system identification experiments. The pendulum-like dynamics model with identified parameters is then linearized for future controller design and validated with experimental data.