-
Towards learning-based planning:The nuPlan benchmark for real-world autonomous driving
Authors:
Napat Karnchanachari,
Dimitris Geromichalos,
Kok Seang Tan,
Nanxiang Li,
Christopher Eriksen,
Shakiba Yaghoubi,
Noushin Mehdipour,
Gianmarco Bernasconi,
Whye Kit Fong,
Yiluan Guo,
Holger Caesar
Abstract:
Machine Learning (ML) has replaced traditional handcrafted methods for perception and prediction in autonomous vehicles. Yet for the equally important planning task, the adoption of ML-based techniques is slow. We present nuPlan, the world's first real-world autonomous driving dataset, and benchmark. The benchmark is designed to test the ability of ML-based planners to handle diverse driving situa…
▽ More
Machine Learning (ML) has replaced traditional handcrafted methods for perception and prediction in autonomous vehicles. Yet for the equally important planning task, the adoption of ML-based techniques is slow. We present nuPlan, the world's first real-world autonomous driving dataset, and benchmark. The benchmark is designed to test the ability of ML-based planners to handle diverse driving situations and to make safe and efficient decisions. To that end, we introduce a new large-scale dataset that consists of 1282 hours of diverse driving scenarios from 4 cities (Las Vegas, Boston, Pittsburgh, and Singapore) and includes high-quality auto-labeled object tracks and traffic light data. We exhaustively mine and taxonomize common and rare driving scenarios which are used during evaluation to get fine-grained insights into the performance and characteristics of a planner. Beyond the dataset, we provide a simulation and evaluation framework that enables a planner's actions to be simulated in closed-loop to account for interactions with other traffic participants. We present a detailed analysis of numerous baselines and investigate gaps between ML-based and traditional methods. Find the nuPlan dataset and code at nuplan.org.
△ Less
Submitted 6 March, 2024;
originally announced March 2024.
-
Reconciling Inconsistent Molecular Structures from Biochemical Databases
Authors:
Casper Asbjørn Eriksen,
Jakob Lykke Andersen,
Rolf Fagerberg,
Daniel Merkle
Abstract:
Information on the structure of molecules, retrieved via biochemical databases, plays a pivotal role in various disciplines, such as metabolomics, systems biology, and drug discovery. However, no such database can be complete, and the chemical structure for a given compound is not necessarily consistent between databases. This paper presents StructRecon, a novel tool for resolving unique and corre…
▽ More
Information on the structure of molecules, retrieved via biochemical databases, plays a pivotal role in various disciplines, such as metabolomics, systems biology, and drug discovery. However, no such database can be complete, and the chemical structure for a given compound is not necessarily consistent between databases. This paper presents StructRecon, a novel tool for resolving unique and correct molecular structures from database identifiers. StructRecon traverses the cross-links between database entries in different databases to construct what we call an identifier graph, which offers a more complete view of the total information available on a particular compound across all the databases. In order to reconcile discrepancies between databases, we first present an extensible model for chemical structure which supports multiple independent levels of detail, allowing standardisation of the structure to be applied iteratively. In some cases, our standardisation approach results in multiple structures for a given compound, in which case a random walk-based algorithm is used to select the most likely structure among incompatible alternates. We applied StructRecon to the EColiCore2 model, resolving a unique chemical structure for 85.11 % of identifiers. StructRecon is open-source and modular, which enables the potential support for more databases in the future.
△ Less
Submitted 24 August, 2023;
originally announced August 2023.
-
SeaShark: Towards a Modular Multi-Purpose Man-Portable AUV
Authors:
Jesper Haahr Christensen,
Marco Jacobi,
Martin Clemmensen Rotne,
Morten Soede Nielsen,
Max Abildgaard,
Claus Eriksen,
Lars Valdemar Mogensen
Abstract:
In this work, we present the SeaShark AUV: a modular, easily configurable, one-man portable micro-AUV. The SeaShark AUV is conceived as modular parts that fit around a central main tube, which holds battery and other vital parts. The head unit comprises easy exhangeable, stackable, and 360 degree rotatable payload sections to quickly obtain a suitable configuration for many objectives. We employ n…
▽ More
In this work, we present the SeaShark AUV: a modular, easily configurable, one-man portable micro-AUV. The SeaShark AUV is conceived as modular parts that fit around a central main tube, which holds battery and other vital parts. The head unit comprises easy exhangeable, stackable, and 360 degree rotatable payload sections to quickly obtain a suitable configuration for many objectives. We employ navigation no better than dead reckoning or relative navigation with respect to some well-known structure, and thus aim at underwater activities that do not require highly accurate geo-referenced data-points. Operating the SeaShark AUV requires only the vehicle itself and a tablet for mission planning and post-mission review. We have built several complete SeaShark systems and have begun exploring the many possibilities and use-cases in both research and commercial use. Here we present a comprehensive overview and introduction to our AUV and operation principles, and further show data examples for experimental operations for shore-to-sea bio-habitat mapping and in-harbor wall and pier inspection
△ Less
Submitted 12 September, 2020;
originally announced September 2020.
-
Vision and Learning for Deliberative Monocular Cluttered Flight
Authors:
Debadeepta Dey,
Kumar Shaurya Shankar,
Sam Zeng,
Rupesh Mehta,
M. Talha Agcayazi,
Christopher Eriksen,
Shreyansh Daftry,
Martial Hebert,
J. Andrew Bagnell
Abstract:
Cameras provide a rich source of information while being passive, cheap and lightweight for small and medium Unmanned Aerial Vehicles (UAVs). In this work we present the first implementation of receding horizon control, which is widely used in ground vehicles, with monocular vision as the only sensing mode for autonomous UAV flight in dense clutter. We make it feasible on UAVs via a number of cont…
▽ More
Cameras provide a rich source of information while being passive, cheap and lightweight for small and medium Unmanned Aerial Vehicles (UAVs). In this work we present the first implementation of receding horizon control, which is widely used in ground vehicles, with monocular vision as the only sensing mode for autonomous UAV flight in dense clutter. We make it feasible on UAVs via a number of contributions: novel coupling of perception and control via relevant and diverse, multiple interpretations of the scene around the robot, leveraging recent advances in machine learning to showcase anytime budgeted cost-sensitive feature selection, and fast non-linear regression for monocular depth prediction. We empirically demonstrate the efficacy of our novel pipeline via real world experiments of more than 2 kms through dense trees with a quadrotor built from off-the-shelf parts. Moreover our pipeline is designed to combine information from other modalities like stereo and lidar as well if available.
△ Less
Submitted 23 November, 2014;
originally announced November 2014.
