The accurate detection and classification of moving objects is a critical aspect advanced driver assistance systems. We believe that by including the object from multiple sensor detections as key component object's representation perception process, we can improve perceived model environment. First, define composite to include class information in core description. Second, propose complete fusion architecture based on evidential framework solve tracking problem integrating uncertainty...

Mobile ground robots operating on unstructured terrain must predict which areas of the environment they are able to pass in order plan feasible paths. We address traversability estimation as a heightmap classification problem: we build convolutional neural network that, given an image representing patch, predicts whether robot will be traverse such patch from left right. The classifier is trained for specific model (wheeled, tracked, legged, snake-like) using simulation data procedurally...

We introduce a general self-supervised approach to predict the future outputs of short-range sensor (such as proximity sensor) given current long-range camera). assume that former is directly related some piece information be perceived presence an obstacle in position), whereas latter rich but hard interpret directly. instantiate and implement on small mobile robot detect obstacles at various distances using video stream robot's forward-pointing camera, by training convolutional neural...

We introduce a novel approach to long-range path planning that relies on learned model predict the outcome of local motions using possibly partial knowledge. The is trained from dataset trajectories acquired in self-supervised way. Sampling-based planners use this component evaluate edges be added tree. illustrate application pipeline with two robots: complex, simulated, quadruped robot (ANYmal) moving rough terrains; and simple, real, differential-drive (Mighty Thymio), whose geometry...

In this paper, we detail a complete software architecture of key task that an intelligent vehicle has to deal with: frontal object perception. This is solved by processing raw data radar and mono-camera detect track moving objects. Data sets obtained from highways, country roads urban areas were used test the proposed method. Several experiments conducted show method obtains better environment representation, i.e., reduces false alarms missed detections individual sensor evidence.

Intelligent vehicle perception involves the correct detection and tracking of moving objects. Taking into account all possible information at early levels task can improve final model environment. In this paper, we present an evidential fusion framework to represent combine evidence from multiple lists sensor detections. Our considers position, shape appearance represent, associate Although our approach takes place level, propose a general architecture include it as part whole solution....

Automatic knowledge grounding is still an open problem in cognitive robotics. Recent research developmental robotics suggests that a robot's interaction with its environment valuable source for collecting such about the effects of actions. A useful concept this process affordance, defined as relationship between actor, action performed by object on which performed, and resulting effect. This paper proposes formalism defining identifying affordance equivalence. By comparing elements two...

Considering perception as an observation process only is the very reason for which robotic methods are to date unable provide a general capacity of scene understanding. Related work in neuroscience has shown that there strong relationship between and action. We believe considering relation action requires interpret terms agent's own potential capabilities. In this paper, we propose Bayesian approach learning sensorimotor representations through interaction represent notion affordance...

Assessing the traversability of rugged terrain is a difficult challenge for legged robots, especially when they implement multiple, distinct gaits. We tackle this problem on k-rock2 amphibious, sprawling gait robot by training gait-dependent estimator. verify that estimator, trained solely procedurally-generated simulated data, approaches outcomes real-world experiments conducted in an indoor motion capture arena using two terrestrial gaits to cross various obstacles. In simulation...

This paper introduces a multimodal approach for reranking of image retrieval results based on relevance feedback. We consider the problem reordering ranked list images returned by an system, in such way that relevant to query are moved first positions list. propose Markov random field (MRF) model aims at classifying initial retrieval-result as or irrelevant; output MRF is used generate new images. The takes into account (1) rank information provided (2) similarities among list, and (3)...

We consider the problem of planning paths on graphs with some edges whose traversability is uncertain; for each uncertain edge, we are given a probability being traversable (e.g., by learned classifier). categorize different interpretations that meaningful mobile robots navigating partially-known environments, which yields formalization; then focus case in true an edge revealed only when agent visits one its endpoints (Canadian Traveller Problem). In this context, design large simulation...

Resumen en: We propose a novel method to re-order the list of images returned by an image retrieval system (IRS). The combines original order obtained ...

Hot-rolling is a metal forming process that produces workpiece with desired target cross-section from an input through sequence of plastic deformations; each deformation generated by stand composed opposing rolls specific geometry. In current practice, the rolling (i.e., stands and geometry their rolls) needed to achieve given final designed experts based on previous experience, iteratively refined in costly trial-and-error process. Finite Element Method simulations are increasingly adopted...