A causal illusion occurs when people perceive a causal relationship between two events that are not contingent on each other. How this illusion varies when people reason diagnostically (i.e., in an effect-to-cause direction) is not well known. Thus, this experiment explored how diagnostic reasoning affects causal illusions. Participants learnt about a illusory cause-effect relationship in which the probability of the cause and the probability of the effect were orthogonally manipulated to be either high or low. Participants learnt either predictively (i.e., cause-to-effect) or diagnostically, and at test had to make two causal judgements that encouraged either predictive (cause-to-effect) or diagnostic (effect-to-cause) reasoning. Diagnostic reasoning at test increased the strength of the cause density bias and decreased the strength of the effect density bias. It also decreased causal ratings, but only after predictive learning. These findings highlight the promise of diagnostic reasoning as a tool to debias causal decision making.

Question Framing Modulates the Cause Density and Effect Density Biases in Causal Illusions

High-quality parent language may support preschoolers’ language development by shaping underlying neurodevelopment. Indeed, more interactive language relates to higher frontal brain activity (Romeo et al., 2018). However, it remains to be explored whether other features of parent language may also be related to brain activity. This study examined whether parental linguistic sophistication is related to children’s spontaneous high-frequency neural activity, as indexed by EEG gamma power. Transcripts of parent-child dyads (N=36, children M=4.1 years old) completing a puzzle task were analyzed for mean length of utterance (measure of grammatical complexity) and number of different words (measure of vocabulary sophistication). We found that higher parental mean length of utterance was related to higher regional EEG gamma power (F(2.61,80.78)=2.92, p=.046), suggesting that more grammatically complex parent language is linked with higher child neural maturity. Future studies should explore the directionality of this finding and further measures of parent language quality.

Sophisticated parent language is linked to high-frequency neural activity in preschoolers

Learning descriptive 3D features is crucial for understanding 3D scenes with diverse objects and complex structures. However, it is usually unknown whether important geometric attributes and scene context obtain enough emphasis in an end-to-end trained 3D scene understanding network. To guide 3D feature learning toward important geometric attributes and scene context, we explore the help of textual scene descriptions. Given some free-form descriptions paired with 3D scenes, we extract the knowledge regarding the object relationships and object attributes. We then inject the knowledge to 3D feature learning through three classification-based auxiliary tasks. This language-assisted training can be combined with modern object detection and instance segmentation methods to promote 3D semantic scene understanding, especially in a label-deficient regime. Moreover, the 3D feature learned with language assistance is better aligned with the language features, which can benefit various 3D-language multimodal tasks. Experiments on several benchmarks of 3D-only and 3D-language tasks demonstrate the effectiveness of our language-assisted 3D feature learning.  Code is available at https://github.com/Asterisci/Language-Assisted-3D.

Language-Assisted 3D Feature Learning for Semantic Scene Understanding

In the last few years we have seen a revolution in maps, passing from 2 dimensional maps to 3D High Definition maps, available and used for diverse applications, from city design and mobility optimization, to environmental evaluation and citizen services. The majority of these HD maps are static maps, with a maintenance cycle of a few months, are more than sufficient for the vast majority of the applications. During the past 4 years, a new application and usage of 3D HD maps have appeared, for the domain of Fully Automated Driving. In the frame of the H2020 AVENUE project, we are deploying L4 Automated mini-buses in several European cities, studying the issues related to a future full scale, commercial deployment of Automated vehicles to public transportation. One of the identified needs and key issue for a viable commercial deployment is the availability of HD 3D maps of the area where the AVs are operating. Creating via scans the 3D HD maps and annotating them for use by AVs, is a costly and time consuming task: a city can easily have 4.000Kms of routes where AVs would be deployed. In addition, each AV manufacturer has its own model, standard and methodology for the city scan and addition of the required annotations. This, fragmentation of and multitude of standards creates a major obstacle and barrier for a commercially viable deployment of AV public transportation services. From one side it creates a customer locking situation to a specific AV manufacturer and from the other side, since the city is alive and changes daily, requires a very short maintenance cycle of less than 24 hours, increasing the costs and requiring complex logistics for the dissemination of the updates. This presentation will pose the problem, present the related issues and ask the audience to come up with viable solutions!!

Deploying Automated Vehicles in the City: What About the Maps?

Advancements in reliable navigation and mapping rest to a large extent on robust, efficient and scalable understanding of the surrounding environment. The success in recent years have been propelled by the use machine learning techniques for capturing geometry and semantics of environment from video and range sensors. I will discuss approaches object detection, pose recovery, 3D reconstruction and detailed semantic parsing using deep convolutional neural networks (CNNs) as well as challenges of deploying these systems in real-world settings. To overcome the need for large amount of labeled data for training object instance detectors we use active self-supervision provided by a robot traversing an environment and generate training examples for learning novel object embeddings from unlabelled data. The object detectors trained in this manner achieve higher mAP compared to off-the-shelf detectors trained on this limited data. While perception and decision making are often considered separately, I will outline few strategies for jointly optimizing perception and decision making algorithms in the context of elementary navigation tasks. The presented explorations open interesting avenues for control of embodied physical agents and general strategies for design and development of general purpose autonomous systems.

Challenges and Opportunities in Autonomous Navigation

In many computer vision applications, the following problem arises: a collection of observations is interpreted as multiple noisy instances of different classes of objects. The observations can be 2D keypoints in an RG B image, 3D points in lidar scans, 2D-2D point correspondences in stereo matching, trajectories in tracking, etc. The classeses of object can be 2D or 3D geometric entitities like lines and planes, or transformations and relations between images. We will review recent techniques, mainly based on the RANSAC approach, for estimation of single and multiple instances of such object and demostrate their performance on a range of practical problems.

Robust Fitting in Computer Vision

Based on a user-study with experts in motion planning algorithms, we propose a taxonomy of explanations, including kinds of questions/explanations/methods related to motion planning, and a set of design considerations for explainable motion planning.

How experts explain motion planner output: a preliminary user-study to inform the design of explainable planners

We present a 3D human activity simulator, ``HOIsim", focusing on Human-Object Interactions (HOIs). Using HOIsim we model two complex activities for ``lunch" and ``breakfast". Furthermore, we introduce methods to meaningfully randomize activity flows and the environment topology, and make a dataset of the 3D pose maps publicly available.

HOIsim: Synthesizing Realistic 3D Human-Object Interaction Data for Human Activity Recognition

Mecanum-wheeled robots have been thoroughly used to automate tasks in many different applications. However, they are usually controlled by neglecting their dynamics and relying only on their kinematic model. In this paper, we model the behaviour of such robots by taking into account both their equations of motion and the electrodynamic response of their actuators, including dry and viscous friction at their shafts. This allows us to design a model predictive controller aimed to minimise the energy consumed by the robot. The controller also satisfies a number of non-linear inequalities modelling motor voltage limits and obstacle avoidance constraints. The result is an agile controller that can quickly adapt to changes in the environment, while generating fast and energy-efficient manoeuvres towards the goal.

Model Predictive Control for a Mecanum-Wheeled Robot Navigating among Obstacles

Trajectory planning with the consideration of obstacles is a classical task in

Mixed-Integer optimization-based planning for autonomous racing with obstacles and rewards

We propose the design of a nonlinear model predictive controller (NMPC) for the

Control of Fixed-Wing UAV Attitude and Speed Based on Embedded Nonlinear Model Predictive Control

We consider an input-constrained differential-drive robot with actuator dynamics.

Underline digital video library

Question Framing Modulates the Cause Density and Effect Density Biases in Causal Illusions

Sophisticated parent language is linked to high-frequency neural activity in preschoolers

Language-Assisted 3D Feature Learning for Semantic Scene Understanding

Deploying Automated Vehicles in the City: What About the Maps?

Challenges and Opportunities in Autonomous Navigation

Robust Fitting in Computer Vision

How experts explain motion planner output: a preliminary user-study to inform the design of explainable planners

HOIsim: Synthesizing Realistic 3D Human-Object Interaction Data for Human Activity Recognition

Model Predictive Control for a Mecanum-Wheeled Robot Navigating among Obstacles

Mixed-Integer optimization-based planning for autonomous racing with obstacles and rewards

Control of Fixed-Wing UAV Attitude and Speed Based on Embedded Nonlinear Model Predictive Control

On MPC without terminal conditions for dynamic non-holonomic robots

Stay up to date with the latest Underline news!

PRESENTATIONS

CONFERENCES

COMPANY

RESOURCES