Subhodip (Subho) Saha
@SAHA_SUBHODIP
Followers
52
Following
162
Media
24
Statuses
65
AI/Robotics @ Dexterity 🚀| MS UMN Twin Cities 👨🎓| IIT Kharagpur 🎓| Machine Learning, GenAI, Computer Vision, Robotics👋
Bengaluru
Joined December 2014
Had an amazing time at @NVIDIARobotics BYOJ Workshop last week in Hyderabad, diving into cutting-edge robotics and vision AI 🚀! Learned about robotics foundational models, simulation softwares, and synthetic data generation powering physical AI. Here are my key takeaways: 🧵
1
0
1
Isaac Sim: https://t.co/7GZ6obJmMn Gr00t foundational model: https://t.co/PoBce22EJY Cosmos dataset generator: https://t.co/Wf19fEB1LK VSS (video search and summarization): https://t.co/yEJzdIapve [7/7]
github.com
Blueprint for Ingesting massive volumes of live or archived videos and extract insights for summarization and interactive Q&A - NVIDIA-AI-Blueprints/video-search-and-summarization
0
0
2
Would like to thank @NVIDIARobotics for sponsoring the event, and @nvidia team, specifically, Ninad Madhab, Vibodh Koushik Haveri, Kabilan Kb for organizing the event. Would be looking forward to next year’s BYOJ! 🌟 What’s your take on where Physical AI is headed? 🤖 [6/7]
1
0
1
🔹 Video Search & Summarization (VSS): Learned about NVIDIA’s VSS Blueprint, which I found particularly interesting. Leveraging VLMs such as Cosmos Reason for contextual video understanding along with LLMs, It introduces a vision AI agent designed to efficiently analyze and
1
0
1
🔹 Dataset Generation: MobilityGen creates robust robot datasets, Cosmos generates varied environments (textures, lighting), and Replicator produces labels – together addressing key bottlenecks in robotics perception. [4/7]
1
0
1
🔹 Robotics Foundation Models: Explored Groot and Groot Mimic, fine-tuned for versatile robot movements, enabling seamless navigation across diverse tasks and scenarios. [3/7]
1
0
1
🔹 Robotics Simulations: Hands-on session with simulation engines such as Isaac Sim, Isaac Lab. The Newton physics engine (developed with NVIDIA, Google, and Disney) brings realistic dynamics to robotics simulations. [2/7]
1
0
1
🔹 Robotics Foundation Models: Explored Groot and Groot Mimic, fine-tuned for versatile robot movements, enabling seamless navigation across diverse tasks and scenarios. [3/7]
0
0
1
🔹 Robotics Simulations: Hands-on session with simulation engines such as Isaac Sim, Isaac Lab. The Newton physics engine (developed with NVIDIA, Google, and Disney) brings realistic dynamics to robotics simulations. [2/7]
1
0
1
Fragile. Oversized. Heavy. Awkward. Expensive. Everything robots are bad at figuring out how to move. Everything Dexterity was built for. Next week: we show you how. #PhysicalAI #Robotics #Automation #AI
1
5
11
🚀 I’m excited to share that I recently joined @DexterityRobots as a Machine Learning Engineer. I’m working on building superhumanoid robots powered by computer vision to develop automation solutions for logistics and warehouses. ✨ It’s amazing to be working with such a talented
0
0
2
Interesting 🌟
BREAKING: Apple just proved AI "reasoning" models like Claude, DeepSeek-R1, and o3-mini don't actually reason at all. They just memorize patterns really well. Here's what Apple discovered: (hint: we're not as close to AGI as the hype suggests)
0
0
1
NVIDIA GR00T foundational model: https://t.co/PoBce22EJY MaskedMimic algorithm: https://t.co/UIixiBoP6U [5/5]
developer.nvidia.com
Creating interactive simulated humanoids that move naturally and respond intelligently to diverse control inputs remains one of the most challenging problems in computer animation and robotics.
0
0
1
GR00T is still in early development, but navigation algorithms like MaskedMimic are clearly accelerating progress. We’ll be getting closer to general-purpose robots that could assist in homes, factories, or even exploration missions. 🧩 Where do you see the next foundational
1
0
1
🎮 I created a simulation in Isaac Lab, the model trained using MaskedMimic algorithm, showing a robot walking on flat terrain, as in the attached demo. I’m also sharing NVIDIA’s demos on robots sitting on sofa, and walking on complex terrain to show what’s possible. [3/5]
1
0
1
🧠 A key part of GR00T’s navigation system is MaskedMimic algorithm. It uses human motion data to predict missing movements—like inferring leg steps from just head/hand positions. Here’s how it works: 1️⃣ First a reinforcement learning model trained on full-body motion tracking
1
0
1
Automation India Expo 2025 (@IEDAutomation) highlighted AI’s expanding role in robotics and machine vision, driving real-time autonomy in industries. 💡 Where do you see AI making the biggest impact in automation? Share your take or DM me if you’re at the intersection of AI and
0
0
1
🌟 Exciting Companies I Met: 🔹 Robotics: @alstrut (with @URobotsAPAC ), @Gridbots, Virya Autonomous Technologies, @ROBONETICS1, @idec, Hachidori Robotics Private Limited 🔹 AI-driven Machine Vision: @_KamerAI, SwitchOn Inc. 🔹 Industry Automation: @Yokogawa, JUMO India Pvt
1
0
1
📷 AI in Machine Vision • Object Detection & Tracking – YOLO & OpenCV-based AI for real-time counting, measuring, and speed detection. 🏭 Big Companies Using AI • Predictive Maintenance & Dashboard Analysis – ML-driven workflow automation for industries. Generative AI and
1
0
1