Skip to content
View bakshienator77's full-sized avatar
10 followers · 3 following

Achievements

Achievement: Pair Extraordinairex4Achievement: StarstruckAchievement: Pull Sharkx3Achievement: QuickdrawAchievement: YOLO

Achievements

Achievement: Pair Extraordinairex4Achievement: StarstruckAchievement: Pull Sharkx3Achievement: QuickdrawAchievement: YOLO

Block or report bakshienator77

Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
bakshienator77/README.md

Nikhil Angad Bakshi

I build learning systems for robots and lead teams that turn research into working products.

I'm a Founding Robotics ML Engineer at Tuesday Labs, where I build reinforcement-learning and robot behavior systems for a home-tidying robot. Previously, at Arena AI, I led development of an AI hardware engineer and built multimodal representation-learning systems. My research at Carnegie Mellon focused on multi-robot search under uncertainty.

Website · LinkedIn · Google Scholar

Selected work

  • Building a home-tidying robot — Tuesday Labs. First full-time hire; built the deep RL stack and led a software team of four through five prototype iterations. Developed locomotion and expressive robot behavior, and reduced hardware-validation cycles from weeks to one day.
  • AI for hardware engineering — Arena AI. Led development of an AI hardware engineer, with end-to-end ownership of a transformer-based multimodal representation-learning pipeline. Project video and announcement.
  • Learning under uncertainty — CMU Robotics Institute. Developed algorithms for decentralized search with heterogeneous teams of robots, connecting probabilistic decision-making with field robotics.

Selected research

GUTS: Generalized Uncertainty-Aware Thompson Sampling for Multi-Agent Active Search
Nikhil Angad Bakshi, Tejus Gupta, Ramina Ghods, Jeff Schneider · ICRA 2023
Multi-agent active search under uncertainty. Winner of the Outstanding Deployed Systems Paper award. The GUTS implementation is included in the STAR repository as a state-of-the-art baseline for comparison.
Paper · Video · Code · Award

Stealthy Terrain-Aware Multi-Agent Active Search (STAR)
Nikhil Angad Bakshi, Jeff Schneider · CoRL 2023
Terrain-aware planning that balances finding targets with remaining concealed.
Paper · Video · Code

M.S. Robotics, Carnegie Mellon University · B.Tech. Mechanical Engineering, minor in Computer Science, IIT Kharagpur.

More publications · Master's thesis

Pinned Loading

  1. Stealthy-Terrain-Aware-Reconnaissance-and-Search Stealthy-Terrain-Aware-Reconnaissance-and-Search Public

    Associated code with S.T.A.R. paper accepted in CoRL 2023

    Python 6 1

  2. AnirudhaRamesh/Learning-to-Detect-by-Learning-to-Predict AnirudhaRamesh/Learning-to-Detect-by-Learning-to-Predict Public

    Detection by Learning to Predict!

    Python 2

  3. e2e_deep_visual_odometry e2e_deep_visual_odometry Public

    My solution for the TartanVO VSLAM challenge Mono Track [https://www.aicrowd.com/challenges/tartanair-visual-slam-mono-track]

    Jupyter Notebook 2 2

  4. pointnet_simplified pointnet_simplified Public

    A detailed analysis of the limitations of a naive implementation of PointNet without the transform blocks with respect to different variations in testing data on the classification and segmentation…

    Python 1

  5. GANs_vs_VAEs_analysis GANs_vs_VAEs_analysis Public

    Examining the stability of training and quality of image generation as well as interpolatability of latent space with three types of GAN training methods and VAEs

    Python 1

  6. vqa_mscoco vqa_mscoco Public

    Uses hierarchical co-attention to models to complete the visual question answering task on a subset of MSCOCO VQA dataset. Analysis and critiques also included.

    Python