Semantic-Based Active Perception for Humanoid Visual Tasks with Foveal Sensors

The aim of this work is to establish how accurately a recent semantic-based foveal active perception model is able to complete visual tasks that are regularly performed by humans, namely, scene exploration and visual search. This model exploits the ability of current object detectors to localize and classify a large number of object classes and to update a semantic description of a scene across multiple fixations. It has been used previously in scene exploration tasks. In this paper, we revisit the model and extend its application to visual search tasks. To illustrate the benefits of using semantic information in scene exploration and visual search tasks, we compare its performance against traditional saliency-based models. In the task of scene exploration, the semantic-based method demonstrates superior performance compared to the traditional saliency-based model in accurately representing the semantic information present in the visual scene. In visual search experiments, searching for instances of a target class in a visual field containing multiple distractors shows superior performance compared to the saliency-driven model and a random gaze selection algorithm. Our results demonstrate that semantic information, from the top-down, influences visual exploration and search tasks significantly, suggesting a potential area of research for integrating it with traditional bottom-up cues.

Pose-free object classification from surface contact features in sequences of Robotic grasps

In this work, we propose two cost efficient methods for object identification, using a multi-fingered robotic hand equipped with proprioceptive sensing. Both methods are trained on known objects and rely on a limited set of features, obtained during a few grasps on an object. Contrary to most methods in the literature, our methods do not rely on the knowledge of the relative pose between object and hand, which greatly expands the domain of application. However, if that knowledge is available, we propose an additional active exploration step that reduces the overall number of grasps required for a good recognition of the object. One of the methods depends on the contact positions and normals and the other depends on the contact positions alone. We test the proposed methods in the GraspIt! simulator and show that haptic-based object classification is possible in pose-free conditions. We evaluate the parameters that produce the most accurate results and require the least number of grasps for classification.

Socially reactive navigation models for mobile robots in dynamic environments

The objective of this work is to expand upon previous works, considering socially acceptable behaviours within robot navigation and interaction, and allow a robot to closely approach static and dynamic individuals or groups. The space models developed in this dissertation are adaptive, that is, capable of changing over time to accommodate the changing circumstances often existent within a social environment. The space model's parameters' adaptation occurs with the end goal of enabling a close interaction between humans and robots and is thus capable of taking into account not only the arrangement of the groups, but also the basic characteristics of the robot itself. This work also further develops a preexisting approach pose estimation algorithm in order to better guarantee the safety and comfort of the humans involved in the interaction, by taking into account basic human sensibilities. The algorithms are integrated into ROS's navigation system through the use of the $costmap2d$ and the $move\_base$ packages. The space model adaptation is tested via comparative evaluation against previous algorithms through the use of datasets. The entire navigation system is then evaluated through both simulations (static and dynamic) and real life situations (static). These experiments demonstrate that the developed space model and approach pose estimation algorithms are capable of enabling a robot to closely approach individual humans and groups, while maintaining considerations for their comfort and sensibilities.

Learning to search for and detect objects in foveal images using deep learning

The human visual system processes images with varied degrees of resolution, with the fovea, a small portion of the retina, capturing the highest acuity region, which gradually declines toward the field of view's periphery. However, the majority of existing object localization methods rely on images acquired by image sensors with space-invariant resolution, ignoring biological attention mechanisms. As a region of interest pooling, this study employs a fixation prediction model that emulates human objective-guided attention of searching for a given class in an image. The foveated pictures at each fixation point are then classified to determine whether the target is present or absent in the scene. Throughout this two-stage pipeline method, we investigate the varying results obtained by utilizing high-level or panoptic features and provide a ground-truth label function for fixation sequences that is smoother, considering in a better way the spatial structure of the problem. Finally, we present a novel dual task model capable of performing fixation prediction and detection simultaneously, allowing knowledge transfer between the two tasks. We conclude that, due to the complementary nature of both tasks, the training process benefited from the sharing of knowledge, resulting in an improvement in performance when compared to the previous approach's baseline scores.

Force Feedback Control For Dexterous Robotic Hands Using Conditional Postural Synergies

We present a force feedback controller for a dexterous robotic hand equipped with force sensors on its fingertips. Our controller uses the conditional postural synergies framework to generate the grasp postures, i.e. the finger configuration of the robot, at each time step based on forces measured on the robot's fingertips. Using this framework we are able to control the hand during different grasp types using only one variable, the grasp size, which we define as the distance between the tip of the thumb and the index finger. Instead of controlling the finger limbs independently, our controller generates control signals for all the hand joints in a (low-dimensional) shared space (i.e. synergy space). In addition, our approach is modular, which allows to execute various types of precision grips, by changing the synergy space according to the type of grasp. We show that our controller is able to lift objects of various weights and materials, adjust the grasp configuration during changes in the object's weight, and perform object placements and object handovers.

3DSGrasp: 3D Shape-Completion for Robotic Grasp

Real-world robotic grasping can be done robustly if a complete 3D Point Cloud Data (PCD) of an object is available. However, in practice, PCDs are often incomplete when objects are viewed from few and sparse viewpoints before the grasping action, leading to the generation of wrong or inaccurate grasp poses. We propose a novel grasping strategy, named 3DSGrasp, that predicts the missing geometry from the partial PCD to produce reliable grasp poses. Our proposed PCD completion network is a Transformer-based encoder-decoder network with an Offset-Attention layer. Our network is inherently invariant to the object pose and point's permutation, which generates PCDs that are geometrically consistent and completed properly. Experiments on a wide range of partial PCD show that 3DSGrasp outperforms the best state-of-the-art method on PCD completion tasks and largely improves the grasping success rate in real-world scenarios. The code and dataset will be made available upon acceptance.

Active Gaze Control for Foveal Scene Exploration

Active perception and foveal vision are the foundations of the human visual system. While foveal vision reduces the amount of information to process during a gaze fixation, active perception will change the gaze direction to the most promising parts of the visual field. We propose a methodology to emulate how humans and robots with foveal cameras would explore a scene, identifying the objects present in their surroundings with in least number of gaze shifts. Our approach is based on three key methods. First, we take an off-the-shelf deep object detector, pre-trained on a large dataset of regular images, and calibrate the classification outputs to the case of foveated images. Second, a body-centered semantic map, encoding the objects classifications and corresponding uncertainties, is sequentially updated with the calibrated detections, considering several data fusion techniques. Third, the next best gaze fixation point is determined based on information-theoretic metrics that aim at minimizing the overall expected uncertainty of the semantic map. When compared to the random selection of next gaze shifts, the proposed method achieves an increase in detection F1-score of 2-3 percentage points for the same number of gaze shifts and reduces to one third the number of required gaze shifts to attain similar performance.

Action-conditioned Benchmarking of Robotic Video Prediction Models: a Comparative Study

A defining characteristic of intelligent systems is the ability to make action decisions based on the anticipated outcomes. Video prediction systems have been demonstrated as a solution for predicting how the future will unfold visually, and thus, many models have been proposed that are capable of predicting future frames based on a history of observed frames~(and sometimes robot actions). However, a comprehensive method for determining the fitness of different video prediction models at guiding the selection of actions is yet to be developed. Current metrics assess video prediction models based on human perception of frame quality. In contrast, we argue that if these systems are to be used to guide action, necessarily, the actions the robot performs should be encoded in the predicted frames. In this paper, we are proposing a new metric to compare different video prediction models based on this argument. More specifically, we propose an action inference system and quantitatively rank different models based on how well we can infer the robot actions from the predicted frames. Our extensive experiments show that models with high perceptual scores can perform poorly in the proposed action inference tests and thus, may not be suitable options to be used in robot planning systems.

Action Anticipation for Collaborative Environments: The Impact of Contextual Information and Uncertainty-Based Prediction

For effectively interacting with humans in collaborative environments, machines need to be able anticipate future events, in order to execute actions in a timely manner. However, the observation of the human limbs movements may not be sufficient to anticipate their actions in an unambiguous manner. In this work we consider two additional sources of information (i.e. context) over time, gaze movements and object information, and study how these additional contextual cues improve the action anticipation performance. We address action anticipation as a classification task, where the model takes the available information as the input, and predicts the most likely action. We propose to use the uncertainty about each prediction as an online decision-making criterion for action anticipation. Uncertainty is modeled as a stochastic process applied to a time-based neural network architecture, which improves the conventional class-likelihood (i.e. deterministic) criterion. The main contributions of this paper are three-fold: (i) we propose a deep architecture that outperforms previous results in the action anticipation task; (ii) we show that contextual information is important do disambiguate the interpretation of similar actions; (iii) we propose the minimization of uncertainty as a more effective criterion for action anticipation, when compared with the maximization of class probability. Our results on the Acticipate dataset showed the importance of contextual information and the uncertainty criterion for action anticipation. We achieve an average accuracy of 98.75% in the anticipation task using only an average of 25% of observations. In addition, considering that a good anticipation model should also perform well in the action recognition task, we achieve an average accuracy of 100% in action recognition on the Acticipate dataset, when the entire observation set is used.

Applying Domain Randomization to Synthetic Data for Object Category Detection

Recent advances in deep learning-based object detection techniques have revolutionized their applicability in several fields. However, since these methods rely on unwieldy and large amounts of data, a common practice is to download models pre-trained on standard datasets and fine-tune them for specific application domains with a small set of domain relevant images. In this work, we show that using synthetic datasets that are not necessarily photo-realistic can be a better alternative to simply fine-tune pre-trained networks. Specifically, our results show an impressive 25% improvement in the mAP metric over a fine-tuning baseline when only about 200 labelled images are available to train. Finally, an ablation study of our results is presented to delineate the individual contribution of different components in the randomization pipeline.