Shurjo Banerjee

PhD student at the University of Michigan.
Interested in deep reinforcement learning, incremental learning and vision-and-language problems.

[Jacob Krantz @jacob__krantz]

Language-guided navigators shouldn’t be carted off to a new environment with each new instruction. Instead, they should persist & improve over time.

We present Iterative Vision-and-Language Navigation to evaluate exactly this!
https://t.co/N8fOkGyCqa
https://t.co/ocAAcT1cYm

[Jason Corso @_JasonCorso_]

Congratulations to Dr. Shurjo Banerjee @BanerjeeShurjo for successfully defending his dissertation "Towards Improving Deep Visio-Linguistic Navigation". Job well done! Kudos! @UMRobotics @michigan_AI

[Jacob Krantz @jacob__krantz]

Announcing the new RxR-Habitat Challenge at the #CVPR21 Embodied AI Workshop!

In RxR-Habitat, agents must navigate continuous environments via natural language instructions.

Website: https://t.co/zaOyTxuUoQ
Code: https://t.co/gPWikGAeNF
Workshop: https://t.co/bjN6GaL0e9

[Shurjo Banerjee @BanerjeeShurjo]

Having worked a lot with Jesse recently, I can confidently say he's going be a wonderful advisor who will put the health, well being and success of his students above all else. I'm quite jealous of those applying in the PhD cycle this year. Don't miss out! @CSatUSC

[Jesse Thomason @_jessethomason_]

I'm building a #RoboNLP group @CSatUSC, if you are interested in lifelong learning through human-robot dialog and connecting language to agent perception and action! https://t.co/SXZrUNNSaI

I'll be "at" @emnlp2020 and @corl_conf next week; please reach out if you'd like to chat.

[Shurjo Banerjee @BanerjeeShurjo]

Joint work with @_jessethomason_ and @ProfJasonCorso

[Shurjo Banerjee @BanerjeeShurjo]

Happy to announce Robot Simultaneous Localization and Mapping with Natural Language (RobotSlang), a benchmark of 169 natural language dialogs for controlling a physical robot which will appear at @corl_conf
Paper: https://t.co/qWmxvKDkXb
Code: https://t.co/sb4s84AOL7

[Shurjo Banerjee @BanerjeeShurjo]

RobotSlang contains 5k utterances and over 1k minutes of robot camera and control streams. By training sim agents for dialog-conditioned navigation behavior from a physical robot, we hope that RobotSlang will facilitate sim-2-real transfer for language-guided robot control.