Filters








2,799 Hits in 5.7 sec

Predicting Unobserved Space For Planning via Depth Map Augmentation [article]

Marius Fehr, Tim Taubner, Yang Liu, Roland Siegwart, Cesar Cadena
2019 arXiv   pre-print
In this work, we present an augmented planning system and investigate the effects of employing state-of-the-art depth completion techniques, specifically trained to augment sparse depth maps originating  ...  On real world MAV data the augmented system demonstrates superior performance compared to a planner based on very dense RGB-D depth maps.  ...  ACKNOWLEDGMENT We would like to thank Helen Oleynikova for her help with the planner, Zachary Taylor for enabling the real world experiments and providing the VI-LiDAR setup and Fangchang Ma for his help  ... 
arXiv:1911.05761v1 fatcat:xbdc46dnnzh6jbx7iiqm5sbaca

Uncertainty-driven Planner for Exploration and Navigation [article]

Georgios Georgakis, Bernadette Bucher, Anton Arapin, Karl Schmeckpeper, Nikolai Matni, Kostas Daniilidis
2022 arXiv   pre-print
, and 2) The effectiveness of our planning module when paired with the state-of-the-art DD-PPO method for the point-goal navigation task.  ...  To this end, we present a novel planning framework that first learns to generate occupancy maps beyond the field-of-view of the agent, and second leverages the model uncertainty over the generated areas  ...  In this paper, we introduce Uncertainty-driven Planner for Exploration and Navigation (UPEN), in which we propose a planning algorithm that is informed by predictions over unobserved areas.  ... 
arXiv:2202.11907v1 fatcat:ldk7abfdfrf3nleulkwrkl2l3i

Footprints and Free Space from a Single Color Image [article]

Jamie Watson, Michael Firman, Aron Monszpart, Gabriel J. Brostow
2020 arXiv   pre-print
However, most methods aim to predict the geometry of surfaces that are visible to the camera, which is of limited use when planning paths for robots or augmented reality agents.  ...  We validate our algorithm against a range of strong baselines, and include an assessment of our predictions for a path-planning task.  ...  Special thanks also to Galen Han and Daniyar Turmukhambetov for help capturing, calibrating and preprocessing our handheld camera footage, and to Kjell Bronder for facilitating the dataset annotation.  ... 
arXiv:2004.06376v1 fatcat:zlzjfupn7bcanhrkkuw3p6fpyq

Footprints and Free Space From a Single Color Image

Jamie Watson, Michael Firman, Aron Monszpart, Gabriel J. Brostow
2020 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)  
However, most methods aim to predict the geometry of surfaces that are visible to the camera, which is of limited use when planning paths for robots or augmented reality agents.  ...  We validate our algorithm against a range of strong baselines, and include an assessment of our predictions for a path-planning task.  ...  Special thanks also to Galen Han and Daniyar Turmukhambetov for help capturing, calibrating and preprocessing our handheld camera footage, and to Kjell Bronder for facilitating the dataset annotation.  ... 
doi:10.1109/cvpr42600.2020.00009 dblp:conf/cvpr/WatsonFMB20 fatcat:izh46mzqxvawpjpris73icuqvq

Learning to Look around Objects for Top-View Representations of Outdoor Scenes [chapter]

Samuel Schulter, Menghua Zhai, Nathan Jacobs, Manmohan Chandraker
2018 Lecture Notes in Computer Science  
But instead of hallucinating RGB values, we show that directly predicting the semantics and depths in the occluded areas enables a better transformation into the top-view.  ...  Crucially, training our model does not require costly or subjective human annotations for occluded areas or the top-view, but rather uses readily available annotations for standard semantic segmentation  ...  This baseline consists of two CNNs, one for in-painting in the RGB space and one for semantic and depth prediction.  ... 
doi:10.1007/978-3-030-01267-0_48 fatcat:erre7xn6cfamnb2vwepld3hwfq

Atlas: End-to-End 3D Scene Reconstruction from Posed Images [article]

Zak Murez, Tarrence van As, James Bartolozzi, Ayan Sinha, Vijay Badrinarayanan, Andrew Rabinovich
2020 arXiv   pre-print
Traditional approaches to 3D reconstruction rely on an intermediate representation of depth maps prior to estimating a full 3D model of a scene.  ...  We present an end-to-end 3D reconstruction method for a scene by directly regressing a truncated signed distance function (TSDF) from a set of posed RGB images.  ...  depth maps for the images.  ... 
arXiv:2003.10432v3 fatcat:g3p5j6sdxbb5lb3aoaeqngxfcy

Learning to Look around Objects for Top-View Representations of Outdoor Scenes [article]

Samuel Schulter, Menghua Zhai, Nathan Jacobs, Manmohan Chandraker
2018 arXiv   pre-print
But instead of hallucinating RGB values, we show that directly predicting the semantics and depths in the occluded areas enables a better transformation into the top-view.  ...  Crucially, training our model does not require costly or subjective human annotations for occluded areas or the top-view, but rather uses readily available annotations for standard semantic segmentation  ...  We balance the loss functions for segmentation and depth prediction via L = λ · L Depth + (1 − λ) · L Seg , where λ ∈ [0, 1].  ... 
arXiv:1803.10870v1 fatcat:4d7zd2nl6ffixkmn4yje7wirjy

Autonomous Navigation in Unknown Environments using Sparse Kernel-based Occupancy Mapping [article]

Thai Duong, Nikhil Das, Michael Yip, Nikolay Atanasov
2020 arXiv   pre-print
We propose a new map representation, in which occupied and free space are separated by the decision boundary of a kernel perceptron classifier.  ...  We demonstrate the effectiveness of our mapping and collision checking algorithms for autonomous navigation of an Ackermann-drive robot in unknown environments.  ...  Assuming unobserved regions are free, we rely onm t k to plan a robot trajectory to a goal region C goal ⊂ C f ree .  ... 
arXiv:2002.01921v1 fatcat:2v2a43gsd5c65fqfkzwvfzv7uq

Autonomous Driving with Deep Learning: A Survey of State-of-Art Technologies [article]

Yu Huang, Yue Chen
2020 arXiv   pre-print
We investigate the major fields of self-driving systems, such as perception, mapping and localization, prediction, planning and control, simulation, V2X and safety etc.  ...  Due to the limited space, we focus the analysis on several key areas, i.e. 2D and 3D object detection in perception, depth estimation from cameras, multiple sensor fusion on the data, feature and task  ...  [169] design Geometric Neural Network (GeoNet) to jointly predict depth and surface normal maps by incorporating geometric relation between depth and surface normal via the depth-to-normal and normal-to-depth  ... 
arXiv:2006.06091v3 fatcat:nhdgivmtrzcarp463xzqvnxlwq

Panoptic Multi-TSDFs: a Flexible Representation for Online Multi-resolution Volumetric Mapping and Long-term Dynamic Scene Consistency [article]

Lukas Schmid, Jeffrey Delmerico, Johannes Schönberger, Juan Nieto, Marc Pollefeys, Roland Siegwart, Cesar Cadena
2022 arXiv   pre-print
However, such environments are inevitably subject to long-term changes, which the map needs to account for.  ...  For robotic interaction in environments shared with other agents, access to volumetric and semantic maps of the scene is crucial.  ...  Lastly, free space submaps are queried before resorting to yet unobserved submaps to predict expectations. Only present, i.e. active or persistent submaps, are counted for evaluation. IV.  ... 
arXiv:2109.10165v2 fatcat:hcpgw3z6pjbj3ijqa4hjr6x7em

Learning Sample-Efficient Target Reaching for Mobile Robots [article]

Arbaaz Khan, Vijay Kumar, Alejandro Ribeiro
2018 arXiv   pre-print
The partially observable planning problem is addressed by splitting it into a hierarchical process.  ...  We use convolutional networks to plan locally, and a differentiable memory to provide information about past time steps in the trajectory.  ...  The memory augmented network scheme uses a value iteration network to compute a plan for the locally observed space.  ... 
arXiv:1803.01846v1 fatcat:3pkbqfczhfcynhzpvjrzhf22lu

Environment Predictive Coding for Embodied Agents [article]

Santhosh K. Ramakrishnan, Tushar Nagarajan, Ziad Al-Halah, Kristen Grauman
2021 arXiv   pre-print
We learn these representations via a zone prediction task, where we intelligently mask out portions of an agent's trajectory and predict them from the unmasked portions, conditioned on the agent's camera  ...  We introduce environment predictive coding, a self-supervised approach to learn environment-level representations for embodied agents.  ...  For depth, we extract features from pre-trained models that predict surface normals and keypoints from depth [67] .  ... 
arXiv:2102.02337v1 fatcat:h2qgl5l7wjbpxjxhh3oovhnd4m

Multi-Modal Geometric Learning for Grasping and Manipulation [article]

David Watkins-Valls, Jacob Varley, Peter Allen
2019 arXiv   pre-print
Tactile information is acquired to augment the captured depth information. The network can then reason about the object's geometry by utilizing both the collected tactile and depth information.  ...  This work provides an architecture that incorporates depth and tactile information to create rich and accurate 3D models useful for robotic manipulation tasks.  ...  However, a convex hull will fill regions of unobserved space.  ... 
arXiv:1803.07671v3 fatcat:japmlj5ctbdv3ir73fg7pri7hm

Where Should I Walk? Predicting Terrain Properties from Images via Self-Supervised Learning

Lorenz Wellhausen, Alexey Dosovitskiy, Rene Ranftl, Krzysztof Tadeusz Walas, Cesar Cadena Lerma, Marco Hutter
2019 IEEE Robotics and Automation Letters  
Ground reaction score prediction images are projected into 3D space using depth information from the depth camera and the pose estimate of the robot odometry.  ...  ACCEPTED JANUARY, 2019 a 2D map of the local ground reaction score as a basis for path planning, exhibit intuitive navigation behavior without manually specifying preferred terrain types.  ... 
doi:10.1109/lra.2019.2895390 fatcat:ljjobcjkrvfsrofm6ipaszoyv4

Depth Field Networks for Generalizable Multi-view Scene Representation [article]

Vitor Guizilini, Igor Vasiljevic, Jiading Fang, Rares Ambrus, Greg Shakhnarovich, Matthew Walter, Adrien Gaidon
2022 arXiv   pre-print
Modern 3D computer vision leverages learning to boost geometric reasoning, mapping image data to classical structures such as cost volumes or epipolar constraints to improve matching.  ...  In this paper, we extend this idea and propose to learn an implicit, multi-view consistent scene representation, introducing a series of 3D data augmentation techniques as a geometric inductive prior to  ...  (c) Depth extrapolation to future timesteps. Images and ground-truth depth maps are displayed only for comparison.  ... 
arXiv:2207.14287v1 fatcat:o77recscebhadf4txcqhxmkese
« Previous Showing results 1 — 15 out of 2,799 results