Ayaan Choudhury

Undergraduate Researcher · Supervisor: Prof. Suril Shah

Developing a 7-DOF tendon-driven robotic hand for dexterous manipulation, integrating mechanical design, reinforcement learning-based control, sim-to-real transfer, and Vision-Language-Action (VLA) grounding for language-conditioned manipulation.

Undergraduate Researcher · Supervisor: Dr. Avinash Sharma

Developing learning-based LiDAR localization frameworks for robust autonomous navigation in sparse outdoor environments, in collaboration with Qualcomm.

• Designed a Z-sliced Multi-planar Bird's-Eye-View (BEV) representation that projects 3D LiDAR scans into 15 discretized horizontal planes at 512×512 grid resolution, and built a Scene Coordinate Regression pipeline using a ResNet-based CNN encoder-decoder with a KL-regularized deterministic latent bottleneck and CBAM attention to regress dense 3D world coordinates, with a compact 16M-parameter model that is 6.5× smaller than leading 3D-convolution baselines.
• Integrated RANSAC-based geometric inference with closed-form SVD (Kabsch) pose recovery to align local LiDAR scans with predicted world coordinates, achieving a state-of-the-art average of 0.91 m translation error and 2.30° rotation error on large-scale outdoor benchmarks, with an optimized end-to-end pipeline running at ~21 ms per frame – a 2.6 to 6.9× speedup over diffusion-based and 3D scene-coordinate-regression baselines.

Undergraduate Researcher · Supervisor: Dr. Riby Abraham Boby

Developed an adaptive peg-in-hole robotic assembly framework using multimodal sensing, contact-aware control, and sim-to-real transfer.

• Integrated real-time spatial feedback from camera streams in MuJoCo to guide manipulator motion, improving approach and alignment accuracy by 15% under varied initial configurations.
• Processed high-frequency force-torque sensor signals at 1000 Hz to detect contact states, guide insertion behavior during precision assembly tasks, and improve assembly execution efficiency, reducing overall cycle time by 20% under uncertain insertion conditions.

Research Intern · Noida, Uttar Pradesh

Developed autonomous navigation capabilities for unmanned aerial vehicles operating in previously unmapped indoor environments.

• Built real-time volumetric occupancy mapping from onboard sensor streams and constructed Euclidean Signed Distance Fields (ESDFs) incrementally from Truncated Signed Distance Fields (TSDFs), reducing per-frame TSDF integration latency by ~50% on average relative to state-of-the-art systems (preprint forthcoming).
• Engineered a ROS 2-based onboard autonomy pipeline integrating mapping, planning, and control, and deployed it on a physical UAV flight computer, validating real-time collision-free trajectory generation through live indoor flights in cluttered, GPS-denied environments.

Computer Vision Intern · Gurgaon, Haryana

• Developed OCR pipelines using OpenCV, Tesseract, Detectron2, PaddleOCR, and LayoutParser for processing health and medical invoices in collaboration with IIT Delhi and AIIMS, achieving 91.7% extraction accuracy.
• Enhanced Lipikaar, an OCR library developed by IIT Delhi for indigenous languages, by fixing bugs, improving character recognition accuracy by 20%, and expanding language-processing capabilities.

Teaching Assistant

• Mentored over 300 sophomore students through weekly laboratory sessions, programming support, and conceptual guidance in core machine learning topics.

Bachelor of Technology in Mechanical Engineering with Interdisciplinary Specialization in Robotics and Mobility Systems · CGPA: 8.88

High School Diploma · AISSCE: 95.6