Search

High accuracy of terrain identification is essential for intelligent control of tracked mobile robots. In this article, a learning-based identification framework is proposed to achieve precise driving torque prediction. Experiments under straight-line and steady-state turning maneuvers were conducted to develop terrain identification in three similar terrains. A multiple deep belief networks is applied as the identification layer with three kinds of signal sources. An equivalent weight algorithm with training experience effectively integrates the results from different signal sources to improve the identification accuracy. With the experiment and identification results, a method combining numerical approximation and Gaussian process (GP) is presented to predict driving torque. A combined Gaussian kernel with long- and short-term characteristics is selected to enhance prediction performance. The results from combined signal sources under straight-line maneuvers yield over 98% accuracy, which exceeds that from other sources. The integration algorithm obviously improves the identification accuracy and stability compared with a single signal source. The influence of window length in GP is explored with hyperparameters and results. The performance of torque prediction in different terrains is analyzed with inner and outer tracks. Compared with existing methods, the results validate the effectiveness and superior performance of the proposed framework. [ABSTRACT FROM AUTHOR]

Terrain identification is essential to autonomous control algorithm development for mobile robots. This article proposes an integrated framework to identify terrain parameters based on inertial, and driving current signals. Multiple sources are combined to reduce the instability caused by single signals. A dynamic model of the track-soil system is established as the theoretical basis of identification. All signals are processed in the time, frequency, and time-frequency domains. The features of each domain are generated by statistical methods. To analyze, and select superior feature categories, a maximum-relevance, and minimum-redundancy criterion based on Pearson's correlation is proposed to evaluate the priority of features. A probabilistic neural network is used to identify the category of terrain. All results are analyzed with two factors, source, and input, to find the most effective rule of the proposed framework. The crossing combination analysis is taken into consideration to explore all potential improvement. The results show that the driving current yields comparative identification accuracy as inertial signals. Compared to the single signal source, the method using the combined signal source can effectively improve the accuracy of terrain identification. [ABSTRACT FROM AUTHOR]

Visual navigation is an essential part of planetary rover autonomy. Rock segmentation emerged as an important interdisciplinary topic among image processing, robotics, and mathematical modeling. Rock segmentation is a challenging topic for rover autonomy because of the high computational consumption, real-time requirement, and annotation difficulty. This research proposes a rock segmentation framework and a rock segmentation network (NI-U-Net++) to aid with the visual navigation of rovers. The framework consists of two stages: the pre-training process and the transfer-training process. The pre-training process applies the synthetic algorithm to generate the synthetic images; then, it uses the generated images to pre-train NI-U-Net++. The synthetic algorithm increases the size of the image dataset and provides pixel-level masks—both of which are challenges with machine learning tasks. The pre-training process accomplishes the state-of-the-art compared with the related studies, which achieved an accuracy, intersection over union (IoU), Dice score, and root mean squared error (RMSE) of 99.41%, 0.8991, 0.9459, and 0.0775, respectively. The transfer-training process fine-tunes the pre-trained NI-U-Net++ using the real-life images, which achieved an accuracy, IoU, Dice score, and RMSE of 99.58%, 0.7476, 0.8556, and 0.0557, respectively. Finally, the transfer-trained NI-U-Net++ is integrated into a planetary rover navigation vision and achieves a real-time performance of 32.57 frames per second (or the inference time is 0.0307 s per frame). The framework only manually annotates about 8% (183 images) of the 2250 images in the navigation vision, which is a labor-saving solution for rock segmentation tasks. The proposed rock segmentation framework and NI-U-Net++ improve the performance of the state-of-the-art models. The synthetic algorithm improves the process of creating valid data for the challenge of rock segmentation. All source codes, datasets, and trained models of this research are openly available in Cranfield Online Research Data (CORD).

Performance in autonomous driven vehicles is susceptible of degradation when traversing different terrains, thus needing motion controllers to be tuned for different terrain profiles. Such tuning stage is a time consuming process for the programmer or operator, and it is often based on intuition or heuristic approaches, and once tuned, the performance of the vehicle varies according to the terrain nature. In this context, we provide a visual based approach to identify terrain variability and its transitions, while observing and learning the performance of the vehicle using machine learning techniques. Based on the identified terrain and the knowledge regarding the performance of the vehicle, our system self-tunes the motion controller, in real time, to enhance its performance. In particular, the trajectory tracking errors are reduced, the control input effort is decreased, and the effects of the wheel-terrain interaction are mitigated preserving the system robustness. The tests were carried out by simulation and experimentation using a robotized commercial platform. Finally, implementation details and results are included in this paper, showing an enhancement in the motion performance up to 92.4% when the highest accuracy of the terrain classifier was 84.3%.

This research suggests a fuzzy-logic based terrain identification method and the smart prosthetic ankle system, which automatically controls its ankle angle, based on the detected terrain environment, to assist comfortable gait performance of transtibial amputee. Suggested terrain identification method uses shank angle from three different stages of the stance phase in gait cycle (foot-flat, heel-strike, and toe-off) as input for the fuzzy-logic calculation, and detects five different terrain environment (flat, up-slope, down-slope, up-stairs, and down-stairs) within a single step of gait. Suggested smart prosthetic ankle system comprises of 1) load-cell to measure GRF (ground reaction force), 2) IMU (inertial measurement unit) sensor to measure shank angle, 3) actuator and four bar-linkage mechanism to control ankle angle accordingly for detected terrain environment, and 4) MCU (microcontroller unit) to carry out calculations and control algorithm for ankle actuation. To verify the accuracy of the terrain identification method of the system, the experiment was conducted, which consisted of four transtibial amputees to walk on five different terrain conditions, and the result has shown 97.5% detection accuracy. Compared to previous studies, our suggested smart prosthetic ankle system, along with its terrain identification algorithm, uses lesser number of sensors and step cycle to accurately detect gait environment, which may lead to providing better gait assistance and practical convenience for transtibial amputees. [ABSTRACT FROM AUTHOR]

The focus of the present research endeavour is to propose a single channel Electromyogram (EMG) signal driven continuous terrain identification method utilizing a simple classifier. An iterative feature selection algorithm has also been proposed to provide effective information to the classifiers. The proposed method has been validated on EMG signal of fifteen subjects and ten subjects for three and five daily life terrains respectively. Feature selection algorithm has significantly improved the identification accuracy (ANOVA, p -value < 0.05) as compared to principal component analysis (PCA) technique. The average identification accuracies obtained by Support Vector Machine (SVM), Linear Discriminant Analysis (LDA) and Neural Network (NN) classifiers are 96.83 ± 0.28%, 97.45 ± 0.32% and 97.61 ± 0.22% respectively. Subject wise performance (five subjects) of individually trained classifiers shows no significant degradation and difference in performance among the subjects even for the untrained data (ANOVA, p -value > 0.05). The study has been extended to dual muscle approach for terrain identification. However, the proposed algorithm has shown similar performance even with the single muscle approach (ANOVA, p -value > 0.05). The outcome of the proposed continuous terrain identification method shows a pronounced potential in efficient lower limb prosthesis control. [ABSTRACT FROM AUTHOR]

In this study, a terrain recognition system is developed by using three dimensional topographic data, two dimensional aerial images and satellite images. For this system, three dimensional Object Knowledge Based Composite Photogrammetry Technology is used. This technology is applied to terrain areas for the first time by using real data. Thus when selecting the fiducial points, important area elements such as buildings and roads are used. Feasibility of the technology is proven for terrain areas which have structures such as buildings and roads, and have only hills and hollows without human made structures.

Terrain identification is essential for the control of robotic transtibial prostheses to realize smooth locomotion transitions. In this paper, we present a real-time fuzzy-logic-based terrain identification method with multisensor fusion. Five locomotion features, including the foot inclination angle at the first strike, the shank inclination angle at the first strike, foot strike sequence, the foot inclination angle at mid-stance, and the shank inclination angle at toe-off, are used to identify different terrains and terrain transitions. These features are measured by the fusion of two triaxis gyroscopes, two triaxis accelerometers, two force sensitive resistors, and a timer, which can be embedded into the prosthesis. Based on these features, a fuzzy-logic-based identification method is proposed to identify five terrains: level ground, stair ascent, stair descent, ramp ascent, and ramp descent. Moreover, a transition constraint function is developed to improve the identification performance. The execution time of the identification method is 0.79 ms \, \pm \, 0.02 ms (mean \, \pm \, standard error of mean) and continuous terrain identification results show that the method can be operated online in real time. The average identification accuracy of 98.74% \, \pm \, 0.32% is obtained from experiments with six able-bodied and three amputee subjects during steady locomotion periods (no terrain transition). In locomotion transition periods, all the eight transitions we studied are correctly identified and the average identification delay is 9.06% \, \pm \, 3.46% of one gait cycle. [ABSTRACT FROM PUBLISHER]

This paper describes a tactile probe designed for surface identification in a context of all-terrain low-velocity mobile robotics. The proposed tactile probe is made of a small metallic rod with a single-axis accelerometer attached near its tip. Surface identification is based on analyzing acceleration patterns induced at the tip of this mechanically robust tactile probe, while it is passively dragged along a surface. A training dataset was collected over ten different indoor and outdoor surfaces. Classification results for an artificial neural network were positive, with an 89.9% and 94.6% success rate for 1- and 4-s time windows of data, respectively. We also demonstrated that the same tactile probe can be used for unsupervised learning of terrains. For 1-s time windows of data, the classification success rate was only reduced to 74.1%. Finally, a blind mobile robot, performing real-time classification of surfaces, demonstrated the feasibility of this tactile probe as a guidance mechanism. [ABSTRACT FROM AUTHOR]

Visual navigation is an essential part of planetary rover autonomy. Rock segmentation emerged as an important interdisciplinary topic among image processing, robotics, and mathematical modeling. Rock segmentation is a challenging topic for rover autonomy because of the high computational consumption, real-time requirement, and annotation difficulty. This research proposes a rock segmentation framework and a rock segmentation network (NI-U-Net++) to aid with the visual navigation of rovers. The framework consists of two stages: the pre-training process and the transfer-training process. The pre-training process applies the synthetic algorithm to generate the synthetic images; then, it uses the generated images to pre-train NI-U-Net++. The synthetic algorithm increases the size of the image dataset and provides pixel-level masks—both of which are challenges with machine learning tasks. The pre-training process accomplishes the state-of-the-art compared with the related studies, which achieved an accuracy, intersection over union (IoU), Dice score, and root mean squared error (RMSE) of 99.41%, 0.8991, 0.9459, and 0.0775, respectively. The transfer-training process fine-tunes the pre-trained NI-U-Net++ using the real-life images, which achieved an accuracy, IoU, Dice score, and RMSE of 99.58%, 0.7476, 0.8556, and 0.0557, respectively. Finally, the transfer-trained NI-U-Net++ is integrated into a planetary rover navigation vision and achieves a real-time performance of 32.57 frames per second (or the inference time is 0.0307 s per frame). The framework only manually annotates about 8% (183 images) of the 2250 images in the navigation vision, which is a labor-saving solution for rock segmentation tasks. The proposed rock segmentation framework and NI-U-Net++ improve the performance of the state-of-the-art models. The synthetic algorithm improves the process of creating valid data for the challenge of rock segmentation. All source codes, datasets, and trained models of this research are openly available in Cranfield Online Research Data (CORD). [ABSTRACT FROM AUTHOR]

• Foot-to-ground angle kinematics changes significantly on different sloped surface. • Most misclassification occurs between level ground and ramp due to similar structure. • Lowest classification accuracy of level ground followed by ramp (ascent and descent) • IR sensor-based locomotion classification is cost effective and efficient approach. This paper deals with the identification of terrain that is crucial to trigger the damping in semi-active lower limb prosthesis. Objective: To identify level ground and ramp terrains using foot-to-ground angle (FGA) measurement. Methods: First, Instrumented shoe for FGA measurement was developed. Next, data collection from able-bodied (n = 5) and amputee (n = 1) participants was carried out. Finally, a comparison of identification accuracy using support vector machine (SVM) and convolution neural network (CNN) algorithms was done. Results: The average classification accuracy obtained for able-bodied participants and amputee is 79.57% ± 20.32% and 73.06% ± 12.70%, respectively using SVM, whereas it is 83.45% ± 14.50% and 80% ± 12.15% respectively using CNN. Our off-line analysis shows that overall, CNN outperformed SVM with an average of 4.86% increment in classification accuracy in able-bodied participants and 9.54% in an amputee. This study introduced a simplified, low-cost method for terrain identification in the prosthetic control application. [ABSTRACT FROM AUTHOR]