3D object detection is an essential task in autonomous driving. Recent techniques excel with highly accurate rates, provided the input data obtained from precise but expensive LiDAR technology. Approaches based on cheaper monocular or stereo imagery have, until now, resulted drastically lower accuracies --- a gap that commonly attributed to poor image-based depth estimation. However, this paper we argue it not quality of its representation accounts for majority difference. Taking inner...

The development of autonomous vehicles for urban driving has seen rapid progress in the past 30 years. This paper provides a summary current state art environments, based primarily on experiences authors 2007 DARPA Urban Challenge (DUC). briefly summarizes approaches that different teams used DUC, with goal describing some challenges faced environments. also highlights long-term research must be overcome order to enable and points opportunities new technologies applied improving vehicle...

Detecting objects such as cars and pedestrians in 3D plays an indispensable role autonomous driving. Existing approaches largely rely on expensive LiDAR sensors for accurate depth information. While recently pseudo-LiDAR has been introduced a promising alternative, at much lower cost based solely stereo images, there is still notable performance gap. In this paper we provide substantial advances to the framework through improvements estimation. Concretely, adapt network architecture loss...

Reliable and accurate 3D object detection is a necessity for safe autonomous driving. Although LiDAR sensors can provide point cloud estimates of the environment, they are also prohibitively expensive many settings. Recently, introduction pseudo-LiDAR (PL) has led to drastic reduction in accuracy gap between methods based on those cheap stereo cameras. PL combines state-of-the-art deep neural networks depth estimation with by converting 2D map outputs inputs. However, so far these two have...

Many applications of stereo depth estimation in robotics require the generation accurate disparity maps real time under significant computational constraints. Current state-of-the-art algorithms force a choice between either generating mappings at slow pace, or quickly inaccurate ones, and additionally these methods typically far too many parameters to be usable on power- memory-constrained devices. Motivated by shortcomings, we propose novel approach for prediction anytime setting. In...

It is increasingly common for computer users to have access several computers on a network, and hence be able execute many of their tasks any computers. The choice which commonly determined by based knowledge speeds each task the current load computer. A number scheduling systems been developed that balance but such tend minimize idle time rather than users. paper focuses benefits can achieved when system considers both availabilities performance SmartNet resource described compared two...

In the domain of autonomous driving, deep learning has substantially improved 3D object detection accuracy for LiDAR and stereo camera data alike. While networks are great at generalization, they also notorious to overfit all kinds spurious artifacts, such as brightness, car sizes models, that may appear consistently throughout data. fact, most datasets driving collected within a narrow subset cities one country, typically under similar weather conditions. this paper we consider task...

This paper presents a novel optimization-based path planner that is capable of planning multiple contingency paths to directly account for uncertainties in the future trajectories dynamic obstacles. addresses particular problem probabilistic collision avoidance autonomous road vehicles are required safely interact, close proximity, with other unknown intentions. The presented utilizes an efficient spline-based trajectory representation and fast but accurate probability bounds simultaneously...

A robot autonomously completes tasks by transforming its body to match capabilities newly encountered environments.

A cooperative tracking approach for uninhabited aerial vehicles (UAVs) with camera-based sensors is developed and verified flight data. The utilizes a square root sigma point information filter, which takes important properties numerical accuracy (square root), (sigma points), fusion ability (information). Important augmentations to the filter are also delayed data, by estimating correlated processes, moving targets, using multiple models in interacting model formulation. final form of...

Abstract Team Cornell's Skynet is an autonomous Chevrolet Tahoe built to compete in the 2007 DARPA Urban Challenge. consists of many unique subsystems, including actuation and power distribution designed in‐house, a tightly coupled attitude position estimator, novel obstacle detection tracking system, system for augmenting estimates with vision‐based algorithms, path planner based on physical vehicle constraints nonlinear optimization routine, state‐based reasoning agent obeying traffic...

Autonomous mobile robots perform many tasks, such as grasping and inspection, that may require complete models of three-dimensional (3-D) objects in the environment. If little or no knowledge about an object is known a priori, robot must take sensor measurements from strategically determined viewpoints order to reconstruct 3-D model object. We propose autonomous reconstruction approach for very general, with assumptions shape size, bounding box predetermined set candidate viewpoints. A...

Advances in perception for self-driving cars have accelerated recent years due to the availability of large-scale datasets, typically collected at specific locations and under nice weather conditions. Yet, achieve high safety requirement, these perceptual systems must operate robustly a wide variety conditions including snow rain. In this paper, we present new dataset enable robust autonomous driving via novel data collection process - is repeatedly recorded along 15 km route diverse scene...

Abstract A guaranteed estimator for a general class of nonlinear systems and on‐line usage is developed analysed. This filter bounds the linearization error, then applies linear set‐membership such that stability guarantees hold systems. tight bound on error found using interval analysis. recursively estimates an ellipsoidal set in which true state lies. General assumptions include use bounded noises twice continuously differentiable dynamics. When system uniformly observable, it proven...

A study on multivehicle trajectory planning for cooperative reconnaissance problems is presented. Specifically, this work develops understanding and insights into how vehicles cooperate in type missions which target information maximized. The performance metric used to guide the cooperation amount of information, defined using Fisher matrix, that sensing gather over their planned trajectory. receding horizon optimal control formulation developed solved trajectories yield maximum information....

Abstract Midway through the 2007 DARPA Urban Challenge, MIT's robot “Talos” and Team Cornell's “Skynet” collided in a low‐speed accident. This accident was one of first collisions between full‐sized autonomous road vehicles. Fortunately, both vehicles went on to finish race collision thoroughly documented vehicle logs. collaborative study MIT Cornell traces confluence events that preceded examines its root causes. A summary robot–robot interactions during is presented. The logs from are used...

This paper develops a probabilistic anticipation algorithm for dynamic objects observed by an autonomous robot in urban environment. Predictive Gaussian mixture models are used due to their ability probabilistically capture continuous and discrete obstacle decisions behaviors; the predictive system uses output (state estimate covariance) of tracking map environment compute probability distribution over future states specified horizon. A splitting method is proposed based on sigma-point...

This paper considers Bayesian data fusion of conventional robot sensor information with ambiguous human-generated categorical about continuous world states interest. First, it is shown that such soft can be generally modeled via hybrid continuous-to-discrete likelihoods are based on the softmax function. A new procedure, called variational importance sampling (VBIS), then introduced to combine strengths Bayes approximations and fast Monte Carlo methods produce reliable posterior estimates...

Distributed data fusion (DDF) is the process whereby a group of agents sense their local environment, communicate with other agents, and collectively try to infer knowledge about particular process. The applications are many include cooperative robots mapping room, unmanned aerial vehicles (UAVs) geolocating moving object on ground, distributed formation space telescopes, people discussing an interesting issue, either in person or online.

Existing approaches to depth or disparity estimation output a distribution over set of pre-defined discrete values. This leads inaccurate results when the true does not match any these The fact that this is usually learned indirectly through regression loss causes further problems in ambiguous regions around object boundaries. We address issues using new neural network architecture capable outputting arbitrary values, and function derived from Wasserstein distance between predicted...

Covers advancements in spacecraft and tactical strategic missile systems, including subsystem design application, mission analysis, materials structures, developments space sciences, processing manufacturing, operations, applications of technologies to other fields.

A generalized planning methodology for satellite clusters is proposed. The utilizes Hamilton ‐ Jacobi‐Bellman optimality (minimum time or minimum fuel ) to generate quickly a set of maneuvers from an initial stable formation e nal formation. Maneuvers are selected the original based on maneuver time, fuel, and collision proximity. calculated by optimizing switch times using realisticset orbital dynamics. algorithm developed be distributed scaleswell as number satellites increases. minimal...

Self-driving cars must detect other traffic participants like vehicles and pedestrians in 3D order to plan safe routes avoid collisions. State-of-the-art object detectors, based on deep learning, have shown promising accuracy but are prone over-fit domain idiosyncrasies, making them fail new environments-a serious problem for the robustness of self-driving cars. In this paper, we propose a novel learning approach that reduces gap by fine-tuning detector high-quality pseudo-labels target -...