David Sierra-Gonzalez

Update May 2023: I have started a new position as Senior Software Engineer in Perception at Oxa Autonomy.

Before joining Oxa I was a research scientist at Inria Grenoble Rhône-Alpes (team CHROMA), France, where I worked on computer vision and machine learning for autonomous driving applications.

Between 2015-2019 I did a PhD in robotics at Inria under the supervision of Christian Laugier and Jilles Dibangoye, with funding from Toyota Motor Europe.

Before Inria, I earned an MSc in Machine Learning from University College London and worked as a research engineer at the Institute of Robotics and Mechatronics of the German Aerospace Center (DLR).

Email  /  Google Scholar  /  Github  /  LinkedIn

profile photo
Research

My general interests are computer vision, machine learning and robotics. More particularly, in the field of autonomous driving, I'm interested in environment perception (detection, tracking, segmentation), scene prediction, and motion planning.

Papers on motion forecasting and motion planning
MultiLane: Lane intention prediction and sensible lane-oriented trajectory forecasting on centerline graphs
David Sierra-Gonzalez, Anshul Paigwar, Ozgur Erkent, Christian Laugier
ITSC, 2022
pdf / video

A Graph Neural Network is used to predict the centerline that the target intends to follow. Based on that, we predict a distribution over potential endpoints, and multiple lane-oriented trajectory realizations.

Human-like decision-making for automated driving in highways
David Sierra-Gonzalez, Mario Garzon, Jilles Dibangoye, Christian Laugier
ITSC, 2019
pdf

POMDP with Monte-Carlo simulations for lane change decision making. The environment predictions are done with our models from 2016-2017.

Modeling driver behavior from demonstrations in dynamic environments using spatiotemporal lattices
David Sierra-Gonzalez, Ozgur Erkent, Victor Romero-Cano, Jilles Dibangoye, Christian Laugier
ICRA, 2018
pdf / video

Inverse Reinforcement Learning to learn a driver behavioral model on a real setting.

Interaction-aware driver maneuver inference in highways using realistic driver models
David Sierra-Gonzalez, Victor Romero-Cano, Jilles Dibangoye, Christian Laugier
ITSC, 2017
pdf / video

A driver behavioral model learned with Inverse Reinforcement Learning is used as a prior to limit the false positives of an autoregressive model in lane change scenarios.

High-speed highway scene prediction based on driver models learned from demonstrations
David Sierra-Gonzalez, Jilles Dibangoye, Christian Laugier
ITSC, 2016
pdf

A driver behavioral model learned with Inverse Reinforcement Learning in simulation is used to predict future road occupancy on a synthetic, simplified highway setting.

Papers on object detection
Frustrum-Pointpillars: A multi-stage approach for 3D object detection using RGB camera and LiDAR
Anshul Paigwar, David Sierra-Gonzalez, Ozgur Erkent, Christian Laugier
ICCV, Workshop on Autonomous Vehicle Vision, 2021
pdf / video / code

Search space reduction for the PointPillars detection architecture based on 2D detections. At the time of submission F-Pointpillars ranked among the top-5 approaches for BEV pedestrian detection on the KITTI dataset.

Leveraging dynamic occupancy grids for 3D object detection in point clouds
David Sierra-Gonzalez, Anshul Paigwar, Ozgur Erkent, Jilles Dibangoye, Christian Laugier
ICARCV, 2020
pdf / video / dataset

We process sequences of point clouds to obtain dynamic occupancy grids, and feed these to a 3D object detection architecture. By leveraging the grid's dynamic information, the bounding box localization and orientation metrics for small objects such as pedestrians in challenging scenarios are improved by 7% and 27%, respectively.

Papers on semantic grids and environment perception
LAPTNet-FPN: Multi-scale LiDAR-aided projective transform network for real time semantic grid prediction
Manuel Diaz-Zapata, David Sierra-Gonzalez, Ozgur Erkent, Jilles Dibangoye, Christian Laugier
ICRA, 2023
pdf / video

Architecture that augments multi-camera feature maps at different scales with depth information from LiDAR to produce bird's eye view semantic grids of the environment.

TransFuseGrid: Transformer-based Lidar-RGB fusion for semantic grid prediction
Gustavo Salazar-Gomez, Anshul Paigwar, Wenqian Liu, Ozgur Erkent, Manuel Diaz-Zapata, David Sierra-Gonzalez, Christian Laugier
ICARCV, 2022
pdf / video

Architecture that fuses multi-camera and LiDAR data at different scales with self-attention to produce bird's eye view semantic grids of the environment.

LAPTNet: LiDAR-Aided Perspective Transform Network
Manuel Diaz-Zapata, Ozgur Erkent, Christian Laugier, Jilles Dibangoye, David Sierra-Gonzalez
ICARCV, 2022.
pdf

Architecture that fuses multi-camera and LiDAR data by projecting the point-cloud onto an intermediate RGB feature map to produce bird's eye view semantic grids of the environment.

Gridtrack: Detection and tracking of multiple objects in dynamic occupancy grids.
Ozgur Erkent, David Sierra-Gonzalez, Anshul Paigwar, Christian Laugier
ICCVS, 2021
pdf

Multi-Object tracking architecture that leverages dynamic occupancy grids to improve the tracking of occluded targets. It outputs a bird's eye view instance segmentation of the scene.

GndNet: Ground plane estimation and point cloud segmentation for autonomous vehicles
Anshul Paigwar, Ozgur Erkent, David Sierra-Gonzalez, Christian Laugier
IROS, 2020
pdf / video / code

Application of a Pillar Feature Encoding module to ground height prediction and point cloud segmentation.

Semantic grid estimation with a hybrid bayesian and deep neural network approach.
Ozgur Erkent, Christian Wolf, Christian Laugier, David Sierra-Gonzalez, Victor Romero-Cano
IROS, 2018
pdf, video

Architecture for semantic grid prediction through fusion of semantically segmented RGB images and Bayesian occupancy grids.

Papers on human-machine interfaces
Stable myoelectric control of a hand prosthesis using non-linear incremental learning
Arjan Gijsberts, Rashida Bohra, David Sierra-Gonzalez, Alexander Werner, Markus Nowak, Barbara Caputo, Maximo Roa, Claudio Castellini
Frontiers in Neurorobotics, 8(8), 2014
pdf / video

Incremental Ridge Regression and an approximation of the Gaussian Kernel known as Random Fourier Features are combined to predict finger forces from myoelectric signals, both finger-by-finger and grouped in grasping patterns.

A virtual piano-playing environment for rehabilitation based upon ultrasound imaging
Claudio Castellini, Katarina Hertkorn, Mikel Sagardia, David Sierra-Gonzalez, Markus Nowak
BioRob, 2014
pdf

Our ultrasound human-machine interface from 2013 is evaluated in a user study with 10 participants; the subjects are asked to play different sequences of keys in a piano.

Ultrapiano: A novel human-machine interface applied to virtual reality
Mikel Sagardia, Katarina Hertkorn, David Sierra-Gonzalez, Claudio Castellini
ICRA, 2014
pdf / video

Application of our ultrasound human-machine interface to remotely control a virtual piano playing environment in real-time.

A realistic implementation of ultrasound imaging as a human-machine interface for upper-limb amputees
David Sierra-Gonzalez, Claudio Castellini
Frontiers in Neurorobotics, 7(17), 2013
pdf / video

Incremental Ridge Regression is used to enable on-the-fly retraining of the ultrasound-driven interface. A quick data collection of zero and maximum fingertip forces suffices to train a system that predicts intermediate force values spanning a range of 20N per finger.

Ultrasound imaging as a human-machine interface in a realistic scenario
Claudio Castellini, David Sierra-Gonzalez
IROS, 2013
pdf

Ridge Regression is used to map a live feed of ultrasound images of the forearm to fingertip forces.

This website uses a template created by Jon Barron.