Background:
Rapid urbanization and climate challenges are stressing global ecosystems. Meeting UN Sustainable Development Goals (SDGs) around clean energy, water management, sustainable agriculture, and smart infrastructure requires intelligent, adaptive, and scalable AI systems that can operate both in the cloud and at the edge.
While foundation models and generative AI provide powerful reasoning and creativity in scientific domains, deploying them in resource-constrained robotics and IoT environments requires addressing key bottlenecks:
Efficient training & distillation of large models for domain-specific applications.
Reliable inference on edge devices with limited compute/memory.
Alignment techniques for explainability, trustworthiness, and human-in-the-loop collaboration.
Multi-modal fusion of vision, speech, and sensor streams for robotics in unstructured environments.
NVIDIA’s ecosystem of GPUs (A100, RTX Pro 6000), Jetson edge devices (AGX Orin, Orin Nano), and AI frameworks provides the building blocks to accelerate such research.
The Challenge:
Design and implement a generative AI-powered robotics/edge solution that addresses a real-world sustainability challenge, by leveraging NVIDIA hardware and SDKs for efficient training, inference, and deployment.
Participants should aim to:
Train or adapt a foundation model (using distillation/modularity) for a chosen domain (e.g., water quality analysis, smart farming, autonomous navigation).
Optimize inference for Jetson or DGX-powered edge hardware.
Incorporate alignment techniques for explainability and trust in decision-making.
Demonstrate multi-modal or RAG-style integration, combining sensor data, text, and vision for actionable intelligence.
Showcase a working prototype that highlights the transition from large-scale model training to efficient, real-time edge deployment.
Example Problem Directions:
Smart Irrigation Robot → Uses multi-modal AI (vision + soil sensors) to optimize water usage in agriculture, running distilled models on Jetson AGX Orin.
Autonomous Water Monitoring Drone → Employs foundation models fine-tuned for environmental analysis to detect pollutants in rivers/lakes, with efficient inference at the edge.
Disaster Response Assistant → Multi-modal edge AI robot that uses generative AI reasoning + RAG to assist rescue teams with real-time decision support in hazardous environments.
Federated Edge Health Monitoring → Wearables and Jetson-based devices using privacy-preserving federated learning for early detection of health anomalies.
Deliverables:
Code + Hardware Prototype demonstrating the use of NVIDIA devices (Jetson/DGX/RTX/A100).
Benchmark Results for training vs inference efficiency (before/after distillation or optimization).
Demonstration Video (5–7 minutes) showing end-to-end system workflow.
Technical Report aligned with NVIDIA Academic Grant template, covering:
Research problem addressed.
Role of generative AI (training, alignment, inference).
NVIDIA hardware/software stack used.
Sustainability/SDG relevance.
Future research impact.
Evaluation Criteria:
Relevance to NVIDIA Research Areas – Generative AI, alignment, robotics, edge AI.
Technical Rigor – Application of distillation, modularity, inference optimization, or federated learning.
Innovation – Novelty in bridging cloud-scale models with edge robotics.
Impact – Relevance to SDG-focused real-world challenges.
Scalability – Potential for future academic research or community deployment.
References:
NVIDIA Jetson Orin Developer Kits
NVIDIA DGX Systems
NVIDIA RAG & Generative AI Resources
NVIDIA Isaac Robotics Platform
⚡ This challenge positions students and researchers to experiment with frontier AI/edge computing, generate publishable outcomes, and align their prototype directly with NVIDIA’s Academic Grant Program focus areas.
Background:
Rapid urbanization and climate challenges are stressing global ecosystems. Meeting UN Sustainable Development Goals (SDGs) around clean energy, water management, sustainable agriculture, and smart infrastructure requires intelligent, adaptive, and scalable AI systems that can operate both in the cloud and at the edge.
While foundation models and generative AI provide powerful reasoning and creativity in scientific domains, deploying them in resource-constrained robotics and IoT environments requires addressing key bottlenecks:
Efficient training & distillation of large models for domain-specific applications.
Reliable inference on edge devices with limited compute/memory.
Alignment techniques for explainability, trustworthiness, and human-in-the-loop collaboration.
Multi-modal fusion of vision, speech, and sensor streams for robotics in unstructured environments.
NVIDIA’s ecosystem of GPUs (A100, RTX Pro 6000), Jetson edge devices (AGX Orin, Orin Nano), and AI frameworks provides the building blocks to accelerate such research.
The Challenge:
Design and implement a generative AI-powered robotics/edge solution that addresses a real-world sustainability challenge, by leveraging NVIDIA hardware and SDKs for efficient training, inference, and deployment.
Participants should aim to:
Train or adapt a foundation model (using distillation/modularity) for a chosen domain (e.g., water quality analysis, smart farming, autonomous navigation).
Optimize inference for Jetson or DGX-powered edge hardware.
Incorporate alignment techniques for explainability and trust in decision-making.
Demonstrate multi-modal or RAG-style integration, combining sensor data, text, and vision for actionable intelligence.
Showcase a working prototype that highlights the transition from large-scale model training to efficient, real-time edge deployment.
Example Problem Directions:
Smart Irrigation Robot → Uses multi-modal AI (vision + soil sensors) to optimize water usage in agriculture, running distilled models on Jetson AGX Orin.
Autonomous Water Monitoring Drone → Employs foundation models fine-tuned for environmental analysis to detect pollutants in rivers/lakes, with efficient inference at the edge.
Disaster Response Assistant → Multi-modal edge AI robot that uses generative AI reasoning + RAG to assist rescue teams with real-time decision support in hazardous environments.
Federated Edge Health Monitoring → Wearables and Jetson-based devices using privacy-preserving federated learning for early detection of health anomalies.
Deliverables:
Code + Hardware Prototype demonstrating the use of NVIDIA devices (Jetson/DGX/RTX/A100).
Benchmark Results for training vs inference efficiency (before/after distillation or optimization).
Demonstration Video (5–7 minutes) showing end-to-end system workflow.
Technical Report aligned with NVIDIA Academic Grant template, covering:
Research problem addressed.
Role of generative AI (training, alignment, inference).
NVIDIA hardware/software stack used.
Sustainability/SDG relevance.
Future research impact.
Evaluation Criteria:
Relevance to NVIDIA Research Areas – Generative AI, alignment, robotics, edge AI.
Technical Rigor – Application of distillation, modularity, inference optimization, or federated learning.
Innovation – Novelty in bridging cloud-scale models with edge robotics.
Impact – Relevance to SDG-focused real-world challenges.
Scalability – Potential for future academic research or community deployment.
References:
NVIDIA Jetson Orin Developer Kits
NVIDIA DGX Systems
NVIDIA RAG & Generative AI Resources
NVIDIA Isaac Robotics Platform
⚡ This challenge positions students and researchers to experiment with frontier AI/edge computing, generate publishable outcomes, and align their prototype directly with NVIDIA’s Academic Grant Program focus areas.