Victoria Dean

Ethics and Robotics (16-735)

Deep Reinforcement Learning for Robotics (16-881)

FIRST Robotics Competition

Introduction to Deep Learning (6.S191)

MIT Global Teaching Labs: Python and Machine Learning

#HelloWorld: Web Development

Code for Good (6.S187)

Introduction to Electrical Engineering and Computer Science (6.01)

Discovery Lab

Train Offline, Test Online: A Real Robot Learning Benchmark

Gaoyue Zhou*, Victoria Dean*, Mohan Kumar Srirama, Aravind Rajeswaran, Jyothish Pari, Kyle Hatch, Aryan Jain, Tianhe Yu, Pieter Abbeel, Lerrel Pinto, Chelsea Finn, Abhinav Gupta

Train Offline, Test Online (TOTO) provides remote users with access to shared robots for evaluating methods on common tasks and an open-source dataset of these tasks for offline training. We present results on TOTO comparing five pretrained visual representations and four offline policy learning baselines, remotely contributed by five institutions. We release the benchmark for additional submissions from any user, enabling easy, direct comparison to several methods without the need to obtain hardware or collect data. CMU, Meta AI, NYU, Stanford, UC Berkeley | ICRA 2023. [website, blue sky paper]

Interesting Object, Curious Agent: Learning Task-Agnostic Exploration

Simone Parisi*, Victoria Dean*, Deepak Pathak, Abhinav Gupta

We propose a paradigm change in the formulation and evaluation of task-agnostic exploration. In this setup, the agent first learns to explore across many environments without any extrinsic goal in a task-agnostic manner. Later on, the agent effectively transfers the learned exploration policy to better explore new environments when solving tasks. In this context, we evaluate several baseline exploration strategies and present a simple yet effective approach to learning task-agnostic exploration policies. CMU and FAIR | NeurIPS 2021. [paper, code]

KitchenShift: Evaluating Zero-Shot Generalization of Imitation-Based Policy Learning Under Domain Shifts

Eliot Xing, Abhinav Gupta, Sam Powers*, Victoria Dean*

We propose a testing protocol for evaluating the generalization of policy learning methods. We implement and evaluate KitchenShift, an instance of our testing protocol that applies domain shifts to a realistic kitchen environment. Using KitchenShift, we evaluate imitation and representation learning methods used in current policy learning approaches and find that they are not robust to minor changes in the environment. [paper]

See, Hear, Explore: Curiosity via Audio-Visual Association

Victoria Dean, Shubham Tulsiani, Abhinav Gupta

We introduce See, Hear, Explore (SHE): a new curiosity formulation using multimodal self-supervision. SHE rewards actions that generate novel associations between different sensory modalities (pixels and sounds). Our results on Atari and AI Habitat show that SHE allows for more exploration, is more sample-efficient, and is more robust to noise compared to existing curiosity baselines on these environments. CMU and FAIR | NeurIPS 2020. [paper, website, video]

CaRL: Combining Imitation Learning with Reinforcement Learning

Used reinforcement learning to improve upon imitation learned policies for trajectory generation, with the goal of both learning from observed behavior and generalizing beyond it. Built upon work presented in this paper. Waymo 2018.

Onboard Text Detection

Developed and deployed an onboard text detection system that allows cars to recognize a set of important words to disambiguate a number of situations. Used an accurate but expensive Google OCR system to train a low-latency onboard model. Waymo 2017. [patent]

Deep Learning for Branch Point Selection in RNA Splicing

Built BRANCHR, a state-of-the-art model for predicting branch point selection, a crucial step in RNA splicing. Deep Genomics | WiML and MLCB (oral) at NeurIPS 2016. [paper]

Predicting the Future: Generative Models for Video

Experimented with CNN models to predict a frame given past frames. Presented poster at SuperUROP poster session. MIT Computer Vision group | EECSCon 2016. [poster]

Automated Search for Lyman Alpha Emitters

Developed pattern matching software to search spectra for distant galaxies. Presented poster at 2013 American Astronomical Society meeting. Intel Science Talent Search semifinalist. UC Santa Cruz | American Astronomical Society Meeting 2012 and 2013. [paper, poster 1, poster 2]