Search

Lower extremity amputees face challenges in natural locomotion, which is partially compensated using powered assistive systems, e.g., micro-processor controlled prosthetic leg. In this paper, a radar-based perception system is proposed to assist prosthetic legs for autonomous obstacle traversing, focusing on multiple-step staircases. The presented perception system is composed of a radar module operating with a multiple-input-multiple-output (MIMO) configuration to localize consecutive stair corners. An inertial measurement unit (IMU) is integrated for coordinates correction due to the angular dis-positioning that occurs because of the knee angular motion. The captured information from both sensors is used for staircase dimensioning (depth and height). A shallow neural network (NN) is proposed to model the error due to the hardware limitations and enhance the dimension estimation accuracy (1 cm). The algorithm is implemented on a microcontroller subsystem of the radar kit to qualify the perception system for embedded integration in powered prosthetic legs., 5 pages, 6 figures, 2 tables

A promising method in the automotive industry to anticipate future customer demands is the concept of planned orders. Due to multi-variant products, changing customer demands, and dynamic environments the process of generating planned orders is challenging. This paper introduces an approach using graphical models to generate planned order suggestions in a multi-variant order management process. Bayesian networks are modelled by learning the structure from different data sources, which enable the possibility to directly sample configuration suggestions. To find an optimized graph structure, a method using hierarchical correlation clustering and reinforcement learning is applied, taking into account technical and sales-operated feasibility constraints. The method has high potential in practical usage and is evaluated by a realworld use case of the Dr. Ing. h.c. F. Porsche AG.

The comparatively poor positioning accuracy of industrial robots limits or even prevents their use in many industrial applications. In contrast to other fields of robotic research, robot accuracy improvement has not been significantly boosted by machine learning-based methods yet. For this reason, we carried out four comprehensive series of measurements using a high-precision laser tracker together with a widely used 6-axis articulated robot. The data will be made publicly available to serve as benchmark data for different techniques. Along with the dataset, this paper introduces a hybrid neural network-based approach to compensate both geometric and non-geometric error sources and increase robot positioning accuracy. We compare our method to previous works and demonstrate advanced results.

In this paper, we introduce a novel learning-based approach for grasping known rigid objects in highly cluttered scenes and precisely placing them based on depth images. Our Placement Quality Network (PQ-Net) estimates the object pose and the quality for each automatically generated grasp pose for multiple objects simultaneously at 92 fps in a single forward pass of a neural network. All grasping and placement trials are executed in a physics simulation and the gained experience is transferred to the real world using domain randomization. We demonstrate that our policy successfully transfers to the real world. PQ-Net outperforms other model-free approaches in terms of grasping success rate and automatically scales to new objects of arbitrary symmetry without any human intervention., Accepted at 2020 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS 2020)

Fast progress in the field of Machine Learning and Deep Learning strongly influences the research in many application domains like autonomous driving or health care. In this paper, we propose a batch-wise regularization technique to enhance the interpretability for deep neural networks (NN) by means of a global surrogate rule list. For this purpose, we introduce a novel regularization approach that yields a differentiable penalty term. Compared to other regularization approaches, our approach avoids repeated creating of surrogate models during training of the NN. The experiments show that the proposed approach has a high fidelity to the main model and also results in interpretable and more accurate models compared to some of the baselines.

Safety in Human-Robot Collaboration (HRC) is a bottleneck to HRC-productivity in industry. With robots being the main source of hazards, safety engineers use over-emphasized safety measures, and carry out lengthy and expensive risk assessment processes on each HRC-layout reconfiguration. Recent advances in deep Reinforcement Learning (RL) offer solutions to add intelligence and comprehensibility of the environment to robots. In this paper, we propose a framework that uses deep RL as an enabling technology to enhance intelligence and safety of the robots in HRC scenarios and, thus, reduce hazards incurred by the robots. The framework offers a systematic methodology to encode the task and safety requirements and context of applicability into RL settings. The framework also considers core components, such as behavior explainer and verifier, which aim for transferring learned behaviors from research labs to industry. In the evaluations, the proposed framework shows the capability of deep RL agents learning collision-free point-to-point motion on different robots inside simulation, as shown in the supplementary video.

One obstacle that so far prevents the introduction of machine learning models primarily in critical areas is the lack of explainability. In this work, a practicable approach of gaining explainability of deep artificial neural networks (NN) using an interpretable surrogate model based on decision trees is presented. Simply fitting a decision tree to a trained NN usually leads to unsatisfactory results in terms of accuracy and fidelity. Using L1-orthogonal regularization during training, however, preserves the accuracy of the NN, while it can be closely approximated by small decision trees. Tests with different data sets confirm that L1-orthogonal regularization yields models of lower complexity and at the same time higher fidelity compared to other regularizers., Comment: 8 pages, 18th IEEE International Conference on Machine Learning and Applications (ICMLA) 2019

This work proposes a new strategy for autonomous navigation of mobile robots in sugarcane plantations based on thermal imaging. Unlike ordinary agricultural fields, sugarcane farms are generally vast and accommodates numerous arrangements of row crop tunnels, which are very tall, dense and hard-to-access. Moreover, sugarcane crops lie in harsh regions, which hinder the logistics for employing staff and heavy machinery for mapping, monitoring, and sampling. One solution for this problem is TIBA (Tankette for Intelligent BioEnergy Agriculture), a low-cost skid-steering mobile robot capable of infiltrating the crop tunnels with several sensing/sampling systems. The project concept is to reduce the product cost for making the deployment of a robot swarm feasible over a larger area. A prototype was built and tested in a bioenergy farm in order to improve the understanding of the environment and bring about the challenges for the next development steps. The major problem is the navigation through the crop tunnels, since most of the developed systems are suitable for open field operations and employ laser scanners and/or GPS/IMU, which in general are expensive technologies. In this context, we propose a low-cost solution based on infrared (IR) thermal imaging. IR cameras are simple and inexpensive devices, which do not pose risks to the user health, unlike laser-based sensors. This idea was highly motivated by the data collected in the field, which have shown a significant temperature difference between the ground and the crop. From the image analysis, it is possible to clearly visualize a distinguishable corridor and, consequently, generate a straight path for the robot to follow by using computationally efficient approaches. A rigorous analysis of the collected thermal data, numerical simulations and preliminary experiments in the real environment were included to illustrate the efficiency and feasibility of the proposed navigation methodology.

Capturing the emotions of humans is of paramount importance in human-machine interaction. Here, emotions are typically extracted from the human’s face recorded in image sequences. In this paper, tracking emotions from images is formulated as Bayesian state estimation problem where the system state represents the valence-arousal space of emotions. Handcrafted image features are first mapped to the valence-arousal space by means of a Gaussian process. To allow dynamic emotion tracking, a Kalman filter is derived, where an inequality constraint on the emotional state is employed in order to avoid a drifting state. Experiments based on two well-known facial expression datasets are performed to demonstrate the performance of the proposed approach.

Feature selection is a dimensionality reduction technique that selects a subset of representative features from high-dimensional data in order to eliminate redundancy. Recently, feature selection methods based on sparse learning have attracted significant attention due to their outstanding performance compared with traditional methods that ignore correlation between features. However, they are restricted by design to linear data transformations, a potential drawback given that the underlying correlation structures of data are often non-linear. To leverage a more sophisticated embedding, we propose an autoencoder-based unsupervised feature selection approach that leverages a single-layer autoencoder for a joint framework of feature selection and manifold learning, with spectral graph analysis on the projected data into the learning process to achieve local data geometry preservation from the original data space to the low-dimensional feature space.

In a surveillance environment with high clutter, finding the correct measurement to track associations becomes extremely important for efficient target tracking. This study offers a novel algorithm to retrodict the data association probabilities at any past time instant, when the batch set of measurements is kept in memory. For the retrodiction procedure, the batch association cost is first written explicitly as a binary integer optimization problem with a quadratic cost function and it is shown that the relaxed form of the problem is convex. From the relaxed problem, a lower bound for the optimal association cost is derived, and this lower bound is used as the data association probabilities pertaining to that selected time instant in the past. Due to its consideration of the batch set of data in a retrospective manner, we will call this algorithm as Retrodictive Probabilistic Data Association, RPDA. For simplification of the mathematical analysis, a single point target with no missing measurements, i.e. P D = 1, is taken into account.

When the techniques of random modulation are used in wideband communication receivers, one can design spectrally shaped sequences that mitigate interferers while preserving messages to reduce distortion caused by amplifier nonlinearity and noise. For sampling rates that are too high for standard modulation, one can instead rely on multi-branch architectures involving multiple modulators working at reduced sampling rates. In this paper, we propose an algorithm to design a set of binary sequences to be used in multi-branch modulation to mitigate a strong interferer while allowing for stable message recovery. The implementation consists of a quadratic program that is relaxed into a semidefinite program combined with a randomized projection. While interferer signals are often modeled as a subspace under the discrete Fourier transform, spectrum leakage occurs when the signal contains so-called off-grid frequencies. The Slepian basis provides a much better-suited representation for such bandlimited signals that mitigates spectrum leakage. We use both representations during the evaluation of our design algorithm, where numerical simulations show the advantages of our sequence designs versus the state of the art.

We present a novel egocentric visual localization algorithm for an indoor navigation system, called PERCEPT -V, which is designed to assist the blind and visually impaired users traveling independently in an unfamiliar indoor space. Through the integration of a background extraction module based on Robust Principle Component Analysis (RPCA) into the localization algorithm, we successfully improve the resilience of camera localization to the presence of crowds in the observed scene. Experiments using datasets of videos containing various levels of crowd activity show that the proposed algorithm can increase prominently the reliability of localization performance.