Michael Everett - Aerospace Controls Laboratory

Michael Everett

Alum

GitHub / Website
mfe [at] mit [dot] edu
31-235C

Education

Ph.D. in Mechanical Engineering, MIT, 2020
S.M. in Mechanical Engineering, MIT, 2017
S.B. in Mechanical Engineering, MIT, 2015

Research Interests

Robust Learning
Collision Avoidance
Motion Planning

Projects

	Evidential Traversability Learning Xiaoyi (Jeremy) Cai, Lakshay Sharma, Michael Everett, 2024 Uncertainty-aware traversability learning and risk-aware navigation in off-road terrain
	Risk-Aware Mapping and Planning Lakshay Sharma, Michael Everett, Donggun Lee, Xiaoyi (Jeremy) Cai, 2023 RAMP: A Risk-Aware Mapping and Planning Pipeline for Fast Off-Road Ground Robot Navigation
	Backward Reachability for Neural Feedback Loops Nicholas Rober, Michael Everett, 2022 This project developed a backward reachability strategy to certify safety for systems controlled by neural networks
	Efficient Learning of Neural Network Policies via Imitation Learning and Tube MPC Andrea Tagliabue, Dong-Ki Kim, Michael Everett, 2022 Use a Robust Tube variant of MPC to efficiently learn Neural Network policies via Imitation Learning.
	Risk-Aware Off-Road Navigation Leveraging Semantics Xiaoyi (Jeremy) Cai, Michael Everett, 2022 Use semantics of the environment to infer terrain traversability based on history of speed data.
	Certified Adversarial Robustness for Deep RL Michael Everett, Björn Lütjens, 2020 This project develops deep RL algorithms that are robust to an adversarial perturbation in the observation space
	Self-Driving Delivery Robot Michael Everett, Justin Miller (Ford), 2019 This project develops planning algorithms to enable autonomous navigation in the "last 100m" for delivery robots
	Socially Acceptable Navigation Michael Everett, Steven Chen, 2019 Collision avoidance algorithm using Deep RL.
	Robust and Interpretable RL for Navigation in Pedestrian Crowds Björn Lütjens, Michael Everett, 2018 Deep neural networks can fail overconfidently on novel observations. This work pioneers a reinforcement learning framework that reasons about the predictive confidence and is more robust to novel observations.