Visweswaran
Baskaran

Building robots that don't embarrass themselves in unstructured environments. Currently working on vision-language models and world models for autonomous navigation at Collaborative Robotics. MSE Robotics, Penn GRASP Lab. Interested in math, films, philosophy, and space. Usually reading something I should've read years ago, surfing, snowboarding, or unsuccessfully trying to become decent at racket sports.

visb@co.bot github linkedin cv

Experience

2025–now
AI Research Engineer
Collaborative Robotics
Santa Clara
Building VLMs and world models for robots operating in unstructured, real-world environments.
2025
Graduate Research Assistant
UPenn JIRL
Philadelphia
Zero-shot semantic navigation under Prof. Antonio Loquercio on the DARPA TIAMAT platform.
2024
AI/ML Intern
Collaborative Robotics
Santa Clara
End-to-end conversational AI for robots — multi-modal vision, embodied task execution, on-device LLMs and custom voice models.
2023–25
Graduate Teaching Assistant
University of Pennsylvania
Philadelphia
TA for Deep Learning (ESE 3060), Advanced ML (CIS 6200), and NLP (CIS 5300).
2022
ML Intern
USC Viterbi — Dynamic Robotics Lab
Los Angeles
LSTM-based online dynamics estimation to improve MPC stability and tracking on loaded quadrupeds. IUSSTF-Viterbi Scholar.
2021–22
Deep Learning Researcher
IIT Madras Robotics Lab
Chennai
DDPG-based deep RL for safe position and force tracking in teleoperated systems under communication delays.
2021
Research Intern
IIT Roorkee — Chakrabarty Lab
India
Novel algorithm for shape formation, path planning, and collision avoidance for multi-agent systems in ROS and Gazebo.
2020–23
Researcher
NIT Trichy — Spider R&D
India
ML and robotics research; mentored 20+ students for national-level competitions.

Selected Projects

Gordy
Gordon Ramsay-inspired robot chef. Gemini for skill routing, ACT policies for food manipulation.
Physical AI · 2026
Gordy
  • End-to-end cooking pipeline using Gemini for task decomposition and skill routing.
  • ACT imitation learning policies trained on food manipulation demonstrations.
Abstract Semantic Navigation
Vision-Language Frontier Maps for zero-shot semantic navigation.
Semantic Navigation · 2025
VLFM
  • Evaluated on Unitree Go2 across Habitat and Isaac Sim platforms.
  • Competed in DARPA TIAMAT; language grounding across unseen environments without task-specific training.
Foundation NeRF
Novel SLAM using NeRF + DINOv2 features for geometric consistency with fewer training epochs.
Geometric SLAM · 2024
Foundation NeRF
  • Integrated DINOv2 semantic features into NeRF-based SLAM for geometry-consistent reconstruction.
  • Reduced training epochs while improving map quality on indoor navigation benchmarks.
VLMaps Navigation
Semantic maps on F1Tenth with LLM-based path planning for conversational navigation.
LLM + Robotics · 2024
VLMaps Navigation
  • Built vision-language maps on F1Tenth hardware for zero-shot semantic goal navigation.
  • LLM-based path planning enables natural language commands to robot navigation goals.
MPC for Loaded Quadrupeds
LSTM online dynamics estimation to improve MPC stability and tracking on legged robots.
Legged Robotics · 2022
Quadruped MPC demo
  • 70% increase in stability, 30% increase in tracking accuracy over baseline MPC.
  • Validated in simulation and hardware on Unitree A1 across walking and trotting gaits.
ASL Interpreter
3D-printed glove with sensor fusion and deep learning for ASL-to-speech on a smartphone.
HCI · 2021
ASL glove demo
  • 60% cost reduction vs. existing solutions at 97% recognition accuracy.
  • Python model + Kotlin Android app; validated on Copelia Sim hardware.

Writing

Writing

Misc

reading
watching
visiting