Research in human-centered AI, deep learning, autonomous vehicles & robotics at MIT and beyond. Also, I teach courses on deep learning.

Connect with me (@lexfridman) on Twitter, LinkedIn, Medium, Instagram, Facebook, YouTube, or join mailing list.

Besides research and teaching, I enjoy playing guitar, practicing jiu jitsu, and engaging in deep meaningful conversations including on podcasts both as a guest (e.g., Joe Rogan Experience part 1 & part 2) and as a host (Artificial Intelligence podcast with, for example, Elon Musk).

Research & Publications

Human Side of Tesla Autopilot: Exploration of Functional Vigilance
Summary: Large-scale study of driver functional vigilance during Tesla Autopilot assisted driving. The central observations in this work is that drivers in this dataset use Autopilot frequently and yet appear to maintain a relatively high degree of functional vigilance.
Arguing Machines: Human Supervision of Black Box AI Systems
Summary: Framework for providing human supervision of a black box AI system that makes life-critical decisions. We demonstrate this approach on two applications: (1) image classification and (2) real-world data of AI-assisted steering in Tesla vehicles.
Human-Centered Autonomous Vehicle Systems
Summary: We propose a set of shared autonomy principles for designing and building autonomous vehicle systems in a human-centered way, and demonstrate these principles on a full-scale semi-autonomous vehicle.
MIT Advanced Vehicle Technology Study
Summary: Large-scale real-world AI-assisted driving data collection study to understand how human-AI interaction in driving can be safe and enjoyable. The emphasis is on computer vision based analysis of driver behavior in the context of automation use.
Cognitive Load Estimation in the Wild
Summary: Winner of the CHI 2018 Honorable Mention Award. We propose two novel vision-based methods for cognitive load estimation and evaluate them on a large-scale dataset collected under real-world driving conditions.
Hacking Nonverbal Communication between Pedestrians and Vehicles in VR
Driver Gaze Region Estimation without Use of Eye Movement
Summary: We propose a simplification of the general gaze estimation task by framing it as a gaze region estimation task in the driving context, thereby making it amenable to machine learning approaches. We go on to describe and evaluate one such learning-based approach.
What Can Be Predicted from 6 Seconds of Driver Glances?
Summary: Winner of the CHI 2017 Best Paper Award. We consider a dataset of real-world, on-road driving to explore the predictive power of driver glances.
DeepTraffic
Summary: Traffic simulation and optimization with deep reinforcement learning. Primary goal is to make the hands-on study of deep RL accessible to thousands of students, educators, and researchers.
Crowdsourced Assessment of External Vehicle-to-Pedestrian Displays
Summary: 30 external vehicle-to-pedestrian display concepts for autonomous vehicles were evaluated. Simple, minimalist displays performed best.
Driver Frustration Detection from Audio and Video in the Wild
Summary: A method for detecting driver frustration from both video and audio streams captured during the driver's interaction with an in-vehicle voice-based navigation system. An interesting observation: smiles are more common in unsatisfied vs satisfied interactions.
Automated Synchronization of Driving Data Using Vibration and Steering Events
Summary: A method for automated synchronization of vehicle sensors using accelerometer, telemetry, audio, and dense optical flow from three video sensors.
Active Authentication on Mobile Devices
Summary: An approach for verifying the identity of a smartphone user with with four biometric modalities. We evaluate the approach by collecting real-world behavioral biometrics data from smartphones of 200 subjects over a period of at least 30 days.
Tesla Vehicle Deliveries and Autopilot Mileage Statistics
Summary: Historical data, current estimates, and projections of Tesla vehicle deliveries and Autopilot miles segmented by Autopilot hardware version.