Multi-modal robot programming with natural language and demonstration is a promising technique for efficient teaching of manipulation tasks in industrial environments. In particular, with modern dual-arm robots designed to quickly take over tasks at typical industrial workbenches, the direct teaching of task sequences hardly utilizes the robots' capabilities. We therefore propose a two-staged approach that combines natural language instructions and demonstration with simultaneous task allocation and motion scheduling based on constraint programming. Instead of providing a task description and demonstrations that are replayed to a large extent, the user describes tasks to be scheduled with all relevant constraints and demonstrates relevant locations relative to workpieces and other objects. With explicitly stated constraints on the partial ordering of tasks, the solver allocates the tasks to the robot arms and schedules them in time while avoiding self-collisions and reducing the makespan in our experiment by 33%. The linguistic concepts of naming and grouping enable systematic reuse of sub-task ensembles. The proposed approach is evaluated with four variants of a gluing use-case from furniture assembly in user studies with ten participants. In these user studies, we observed a speed-up for the task definition of more than six times compared to a textual specification of the planning problems using the Python-based planner API.

Language has evolved over centuries and was gradually enriched and improved. The question, how people find assignment between meanings and referents, remains unanswered. There are many of computational models based on the statistical co-occurrence of meaning-reference pairs. Unfortunately, these mapping strategies show poor performance in an environment with a higher number of objects or noise. Therefore, we propose a more robust noise-resistant algorithm. We tested the performance of this novel algorithm with simulated and physical iCub robots. We developed a testing scenario consisting of objects with varying visual properties presented to the robot accompanied by utterances describing the given object. The results suggest that the proposed mapping procedure is robust, resistant against noise and shows better performance than one-step mapping for all levels of noise in the linguistic input, as well as slower performance degradation with increasing noise. Furthermore, the proposed procedure increases the clustering accuracy of both modalities.

This paper introduces a dataset for training and evaluating methods for 6D pose estimation of hand-held tools in task demonstrations captured by a standard RGB camera. Despite the significant progress of 6D pose estimation methods, their performance is usually limited for heavily occluded objects, which is a common case in imitation learning, where the object is typically partially occluded by the manipulating hand. Currently, there is a lack of datasets that would enable the development of robust 6D pose estimation methods for these conditions. To overcome this problem, we collect a new dataset (Imitrob) aimed at 6D pose estimation in imitation learning and other applications where a human holds a tool and performs a task. The dataset contains image sequences of nine different tools and twelve manipulation tasks with two camera viewpoints, four human subjects, and left/right hand. Each image is accompanied by an accurate ground truth measurement of the 6D object pose obtained by the HTC Vive motion tracking device. The use of the dataset is demonstrated by training and evaluating a recent 6D object pose estimation method (DOPE) in various setups. The dataset and code are publicly available at http://imitrob.ciirc.cvut.cz/imitrobdataset.php .

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 gripper morphology and action parameters were the main source of variance, making generalization across embodiment or grip configurations very difficult. 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 data set with 24,000 measurements is publicly available.