Autonomous Multidrone Planning
A deep deconstruction of my M.Tech Thesis: Trajectory planning for autonomous swarms in unknown, partially observable environments.
Interactive 3D WebGL Pathfinder Simulation
Intellectual Property: Indian Patent Office (IPR)
Patent Application No: 202641065450 (Filed May 24, 2026)
"Method and System for Secure Multi-UAV Autonomous Navigation Utilizing Dual-Purpose Segment Trees and Blockchain Consensus"
Thesis: Secure Autonomous Trajectory Planning
To enable unmanned aerial vehicles (UAVs) to navigate safely and autonomously, I designed PEDRA (Programmable Engine for Drone Reinforcement Learning Applications), a training engine that bridges Python scripts and the Unreal Engine simulator using the AirSim API. This setup allows training and testing simulated UAV models (ARDrone, DJIMavic, DJIPhantom) in complex 3D virtual environments under various modes: training, inference, and manual navigation.
Solving Exploration Latency via 2D Segment Trees
Autonomous drones in unknown terrain face a severe exploration versus exploitation dilemma. Dones parse environment coordinates classified as safe/unsafe using simulated LIDAR sensors. However, exploring a massive 3D coordinate array step-by-step is computationally prohibitive.
To optimize this search space, I implemented a 2D Segment Tree data structure. This tree enables dynamic range querying of safe space corridors. A 2D segment tree representing an grid \(N \times M\) allows spatial queries in:
Modification queries are performed in \(O(2 \times n \times \log n \times \log m)\). This spatial range representation reduced the Reinforcement Learning (PPO) search phase from over 150,000 episodes to under 500 episodes, representing a major optimization.
Swarm Security: Decentralized Blockchain Ledger
To secure inter-drone telemetry and path sharing from unauthorized modification or spoofing, I integrated a 6-layer blockchain framework. Transaction payloads are validated and shared across P2P swarm channels.
| Layer | Drone Swarm Implementation | Cryptographic / Network Mechanism |
|---|---|---|
| Data Layer | Telemetry transactions, explored coordinates, and path histories. | SHA-256 block hashing and transaction verification via Merkle tree roots. |
| Network Layer | Swarm nodes broadcast path matrices. | P2P networking. Nodes use asymmetric key pairs (RSA/ECC) to authenticate sender nodes. |
| Consensus Layer | Collision avoidance maps coordination agreement. | Distributed consensus algorithms to approve spatial terrain updates. |
| Incentive Layer | Exploration coordination. | Priority scoring for drone nodes contributing telemetry coordinates in unknown quadrants. |
| Contract Layer | Autonomous path handovers. | Smart contracts enforcing interaction policies and path reservation protocols. |
| Application Layer | Swarm interface dashboards. | Auditable telemetry streams and flight logs. |
Active R&D: AI Agent Protocols & Resource Discovery (ARD)
With the rapid expansion of agentic systems, standardizing how models find, coordinate, and communicate with each other is critical. My active research aligns with the latest frontier protocols:
"The 'Agentic Web' just got its HTTP moment." — Read my detailed industry deconstruction of Agentic Resource Discovery (ARD), Federated Registries, and Google's integration on Gemini Enterprise Agent Platforms.
[View Original Publication on LinkedIn]Agentic Resource Discovery (ARD)
Google's recently announced Agentic Resource Discovery (ARD) specification represents the "HTTP moment" for the Agentic Web. By using standardized Catalogs and Registries, ARD enables Gemini and enterprise agent platforms to query and invoke capabilities dynamically across independent networks without manual integration code.
The Multi-Protocol Agent Stack
I construct architectures around the six core developer protocols: MCP (data/tool access), A2A (inter-agent coordination), UCP (modular commerce), AP2 (cryptographic spending guardrails), A2UI (declarative UI composition), and AG-UI (SSE stream standardization).
Industry Publications & Hackathon Milestones
Detailed technical analyses, engineering deconstructions, and competitive prototype awards shared with the community.
The "Agentic Web" just got its HTTP moment
Deconstruction of Google's Agentic Resource Discovery (ARD) protocol, Federated Registries, and Gemini Enterprise Platforms.
AI for Bharat: Prototype Developer
Shortlisted as a Prototype Developer out of thousands of innovators to deploy serverless agent swarms using Amazon Bedrock, Lambda, API Gateway, and S3 backend.
Top 10 Solo at Google Gen AI Hackathon
Ranked in the top 10 teams as a solo participant at Leela Palace Bengaluru. Built resilient, self-correcting end-to-end agents utilizing Gemini API and Vertex AI.
Gemma 3 Wins Kaggle Bronze Medal
"Teach Gemma 3 to Reason" notebook hit 1,000+ views. Unlocked JAX/Flax CUDA mapping for Group Relative Policy Optimization (GRPO) on local GPUs.
Google Cloud AI Labs Certification
Deep-dive certification on Vertex AI Agent Engine, Agent Development Kit (ADK), Model Context Protocol (MCP), and A2A Protocol swarms.
MedGemma Agentic Starter Kit
A consolidated implementation of the MedGemma Swarm Architecture, merging baseline safety/quantization with FAISS RAG and consensus self-correction.