NVIDIA Inception member Applied research targeting commercial deployment. Trained on NVIDIA hardware. Seeking seed funding.

Physical AI Research Lab

From simulation to real hardware.

Two parallel research tracks at Trigunaï Innovations. Reinforcement-learning policies for drone cinematography, trained in simulation and validated in VR before they touch real hardware. Additive attention guidance for optical see-through AR, designed for the next generation of always-on AR glasses.

PPO cinematography policy trained in NVIDIA Isaac Sim, rendered through OVRTX, validated in our Meta Quest 3 VR application. Same trained ONNX targets a Modal AI Starling 2 autonomous drone next.

What we're building

Two parallel tracks.

Each track has one founder dedicated full-time. Each has working artifacts you can watch today. Each is gated on a single specific hardware purchase to move from simulation to the real world.

Track 01 · Robotics Sim working · hardware next

AI Cinematographer Drone

Lead: Deepak Kumar · 100% dedicated

A PPO policy that films a dancer cinematically — trained for 2000+ epochs on 4096 parallel environments in NVIDIA Isaac Lab on an A10G GPU. Exported to ONNX (~80 KB), runs on-device on Meta Quest 3 at 50 Hz via Unity Inference Engine. Same trained ONNX is designed to deploy directly to a Modal AI Starling 2 autonomous drone with a GoPro Hero 12 — sim-to-real validated in VR first, so we never spend hardware money on an untested policy.

TrainingNVIDIA Isaac Sim / Isaac Lab · A10G GPU
AlgorithmPPO · rl_games
InferenceONNX · Unity Inference Engine · 50 Hz on-device
ValidationQuest 3 VR application · live-rendered drone POV
Hardware targetModal AI Starling 2 · GoPro Hero 12
GatingDrone procurement + 6 mo flight-test runway
Track 02 · Optical AR Design spec · prototyping

ModusXR — Focus Modes for Optical See-Through AR

Lead: Avinash · 100% dedicated

Optical see-through displays are additive — they can emit light but cannot subtract it. The standard "vignette to darken the periphery" attention pattern from VR is physically impossible. ModusXR specifies three additive equivalents — Focal Border, Attention Spotlight, Peripheral Parking — designed for XREAL Project Aura, with an evaluation harness to compare them against pre-declared success criteria rather than aesthetic preference. Currently prototyping on Meta Quest 3 video passthrough as an Aura-honest simulator.

ProgramGoogle Android XR Catalyst (participant)
Target hardwareXREAL Project Aura · optical see-through
Prototype hardwareMeta Quest 3 · video passthrough simulator
Variants3 additive focus modes · evaluation harness
StageDesign spec complete · variant build in progress
GatingXREAL Aura developer kits + sustained engineering
Android XR Catalyst program  →

The connection to commercial

Research, with revenue underneath.

A research lab without revenue is a runway race. A research lab with revenue underneath is an asset that compounds. Trigunaï Innovations runs both: a courses business that funds the research, and a research pipeline that generates the curriculum.

How the courses business and the research feed each other.

Trigunaï Innovations runs a free YouTube series on the universal mind ↔ machine mirror, plus paid live cohorts at trigunai.com. One of the four courses is "Physical AI — Train a Robot in Simulation" — and students in that cohort work inside our actual drone training pipeline as they learn. They get a real RL project on real NVIDIA hardware; we get research labor as part of the curriculum.

This is intentional, not accidental. The Robotics course's syllabus mirrors the drone policy's training pipeline step-by-step. Students who complete the course can credibly say they trained a real policy on real hardware. We get cohort revenue covering the sustained engineering cost of the research, and a steady stream of trained collaborators.

See the courses business → trigunai.com

Stack

NVIDIA from training to deployment.

Both research tracks run on NVIDIA infrastructure end-to-end. Training in Isaac Sim / Isaac Lab on A10G GPUs. Inference on Quest 3 today, Jetson-class hardware tomorrow. The trained ONNX models port across hardware without rewriting the policy.

Simulation
NVIDIA Isaac Sim 6.0 · Isaac Lab 3.0 · OVRTX rendering
Training compute
A10G GPU · 4096 parallel envs · rl_games PPO
Runtime inference
ONNX · Unity Inference Engine 2.5 · 50 Hz on-device
Validation
Meta Quest 3 VR app · Unity 6 · OpenXR · custom shaders
Mocap pipeline
Quest 3 OVRBody · 84 joints @ 60 Hz · pose.bin
Hardware target (drone)
Modal AI Starling 2 · Snapdragon 865 NPU · GoPro Hero 12
Hardware target (AR)
XREAL Project Aura · optical see-through
Cloud + delivery
Azure Container Apps · Postgres · CDN · ACS email

Team

Two engineers, two tracks.

Each founder leads one research track full-time. No employees yet. Equal billing, equal commitment.

Avinash

Co-founder, CTO · Patna

100% on ModusXR / Android XR Catalyst

Leads XR engineering. Shipped TrigunAI's GuruLok meditation VR application live on the Meta Quest store. Built TrigunAI Studio's Quest 3 application end-to-end (Unity, OpenXR, custom shaders, Sacred Forest spawn world, in-headset Cinematographer Studio, Quest-to-cloud live streaming pipeline). Currently dedicated full-time to ModusXR — Google's Android XR Catalyst participant working on additive focus modes for XREAL Project Aura.

GitHub: franka-peg-in-hole GitHub: franka-cube-stacking avinash@trigunai.com

Deepak Kumar

Co-founder, CEO · Mumbai

100% on Drone Policy Training

Leads company direction and the AI cinematographer training pipeline — NVIDIA Isaac Lab PPO training, ONNX export, Unity Inference Engine integration, the v119/v120 production demo embedded above. Background: AI / agentic systems, spatial computing infrastructure, content production workflows. Built TrigunAI's courses platform end-to-end (the cohort revenue that funds this research).

Courses platform: trigunai.com deepak@trigunai.com

Seeking seed funding

Help us move from simulation to real hardware.

Both research tracks have working software artifacts you can watch above. Both are gated on a specific hardware purchase to bridge from simulation to physical deployment — a Modal AI Starling 2 drone for the cinematography policy, and XREAL Aura developer kits for ModusXR. We have a courses business with paying students underneath the research, NVIDIA Inception membership, and two engineers dedicated full-time. Talk to us.

Investor enquiry → email Deepak About NVIDIA Inception →

Open to angel, seed, and strategic conversations. Reply within 24 hours.