Smart Satellites, Smarter Systems: AI Across the Space Ecosystem | AI Impact Summit 2026

Conference Talk Summary

---

Key Topics Covered

• AI-Powered Space Robotics: ISRO's Vomitra humanoid (35 degrees of freedom) for crewed missions and future space stations
• Autonomous Planetary Landing: AI models for safe lunar and Mars landings using simulated environments and transfer learning
• Edge AI & On-Board Computing: Deploying ML models directly on satellites to reduce data transmission, power consumption, and latency
• Remote Sensing & Data Processing: AI techniques for handling massive volumes of Earth observation data (85+ TB/day from NISR)
• Agricultural Applications: Space-based AI for precision farming, weather prediction, pest detection, and carbon credit monitoring
• Guidance, Navigation & Control (GNC): Autonomous spacecraft docking, collision avoidance, and real-time decision-making under communication latency
• Natural Language Processing (NLP): Using speech and context-aware NLP to humanize space assistant/companion robots
• Policy & Funding: India's new AI-in-Space seed fund scheme (grants up to 1 crore per startup)

---

Key Points & Insights

1. Edge Computing Reduces Data Burden: Models can be reduced from 300–800 MB to 30 MB (10x reduction) via optimization on ground before deployment, while maintaining accuracy. This yields 10x faster processing, 5x reduction in model size, and 3x more affordable data delivery.

2. Physics-Informed Approaches Remain Essential: Strong fundamentals in physics, mathematics, and domain-specific knowledge are not replaced by AI—they become more critical. AI amplifies the advantage of those with deep fundamentals (10x better outcomes vs. those without). Physics-Informed Neural Networks (PINNs) validate AI decisions.

3. Simulated Environments Enable Training: Lack of real lunar/Mars surface data is overcome through high-fidelity digital twins and simulated environments (e.g., modeled Apollo landing sites), where AI models can be trained and validated before actual missions.

4. Explainability & Robustness Are Non-Negotiable: Control engineers demand guarantees: stability margins, robustness under uncertainty, reproducibility, and the ability to explain why an AI system made a decision. Monte Carlo simulations verify performance across non-nominal conditions.

5. Communication Latency Demands Autonomy: Mars missions face 20-minute round-trip communication delays; Moon missions have 2+ second delays. Onboard autonomous decision-making (millisecond response times) is essential for spacecraft survival, not optional.

6. Data Quality Trumps Volume: With 85 TB/day incoming from NISR, the bottleneck is not quantity but quality and heterogeneity. Standardized, homogenized multimodal data (optical, SAR, drone) is prerequisite for effective AI.

7. Cautious, Incremental Integration: ISRO is introducing autonomy step-by-step (ground operations first, then validated subsystems, then onboard). Parallel development—AI-driven learning alongside mathematical formulations—reduces risk while building confidence.

8. Multi-Disciplinary Teams Accelerate Development: Rather than requiring every engineer to master all domains, multidisciplinary teams (coding, comms, mechanical, controls) reduce learning overhead and allow focus on domain-specific contribution.

9. Over-the-Air Model Updates Are Feasible: Satellite AI models can be patched and updated remotely (similar to phone app updates) to account for geopolitical shifts, climate change, cloud cover, volcanic ash, and disaster management priorities—provided rigorous validation occurs on ground first.

10. New Market Opportunities Emerging: Carbon credit monetization for farmers (via space data and AI), edge AI startups, robotic systems, and space applications are creating job opportunities and lowering barriers to entry for young engineers.

---

Notable Quotes or Statements

• Dr. Vinod Kumar (Opening Remarks): "The final frontier is no longer the destination of human bravery. It has become the ultimate test bed for artificial intelligence."

• Dr. Vinod Kumar (on Space Vision 2047): "The integration of AI into the system is not simply an upgrade. It is a backbone of our vision Space Vision 2047 Vit Goran."

• Dr. Deepti Patil (on AI-enabled lunar landing): "The hardest challenge is taking the real-time decision in an uncertain environment... AI can help us learn the things without knowing the planetary conditions. It may be a case study on the lunar but it can be applicable to other missions also, say Mars."

• Ishita Ganju (on fundamentals): "Fundamentals have to be strong... asking the right questions is tougher than answering them. If you have the right fundamentals you can ask the right questions to the AI and get much more out of it."

• Ashutosh Gupta (on mathematical precursors to AI): "We definitely are using mathematical tools like optimization and regression which are precursors to AI, the foundation on which AI is built."

• Mr. Vinesima (Sky-Serve, on data reduction): "We do 10x reduction, we are 10x faster, 5x reduction in size, and 3x more affordable vis-à-vis data transfer."

• Jagrati Dawas (on real-world impact): "Every day just by sitting at your land you can get to know how many carbon credits can be generated from one hectare of land and how much income I can generate just by signing up."

• Dr. Vinod Kumar (Closing Remarks): "Beyond earth is no longer a distance measured in kilometers but in milliseconds of latency and terabytes of local insights... The frontier is open and it is intelligence."

---

Speakers & Organizations Mentioned

Speakers

• Dr. Vinodkumar — Director, Indian National Space Promotion and Authorization Center (INSPACE); Session Chair & Moderator
• Mr. Duray Raj — AI & Robotics Expert, ISRO; Developer of Vomitra humanoid
• Dr. Deepti Patil — Dean, MKKS Cummins College of Engineering; Autonomous lunar/Mars landing researcher
• Mr. Vinesima — Founder & CEO, Skyserve; Edge AI for geospatial models
• Mr. Ashutosh Gupta — Expert, Space Application Center, Ahmedabad; Remote sensing & data processing
• Ms. Jagrati Dawas — CEO & Founder, ARMS for; Space applications and agricultural insights
• Ms. Ishita Ganju — Senior Scientist, ISRO; Guidance, Navigation & Control (GNC) and autonomous missions

Organizations

• ISRO (Indian Space Research Organisation) — Primary government space agency; Vomitra, Gaganyan, lunar landing, space stations
• INSPACE (Indian National Space Promotion and Authorization Center) — Organizing entity; administers AI-in-Space seed fund
• Skyserve — Private company; edge AI for satellite-based geospatial intelligence
• ARMS for — Private company; space data applications and insights for agriculture
• Space Application Center, Ahmedabad — ISRO facility; remote sensing and Earth observation
• MKKS Cummins College of Engineering — Academic institution; autonomous landing research
• NASA, Canadian Space Agency — Referenced for comparison (Canadian arm on ISS)

---

Technical Concepts & Resources

AI/ML Techniques

• Spiking Neural Networks (SNNs) — Low-power neural networks for edge deployment on FPGAs
• Physics-Informed Neural Networks (PINNs) — Neural networks constrained by physics laws for interpretability
• Transfer Learning — Adapting pre-trained models to space-specific datasets with limited data
• Reinforcement Learning — Policy-based decision-making with physics-informed rewards
• Large Language Models (LLMs) & Transformer Models — For NLP, text generation, and speech understanding
• Monte Carlo Simulations — Statistical validation across non-nominal conditions and worst-case scenarios
• Data-Driven Controls — Combining classical control theory with data-driven learning
• Foundational Models — Large-scale, fine-tunable models replacing domain-specific models

Key Challenges Addressed

• Edge Computing on Spacecraft — Running AI models with limited compute power, memory, and no cloud/internet access
• Training Data Scarcity — Limited space-domain datasets; overcome via simulation and digital twins
• Real-Time Autonomy — Millisecond response times required (communication latency: Moon 2+ sec, Mars 20 min)
• Model Size Reduction — 10x compression (300–800 MB → 30 MB) without accuracy loss
• Explainability & Validation — Ensuring AI decisions are interpretable and robustly tested before deployment

Missions & Systems

• Vomitra Humanoid — 35 degrees of freedom; first unmanned Gaganyan (human spaceflight precursor) mission
• Bharatiya Antariksh Station (space station) — Future platform for autonomous robotic systems
• Autonomous Lunar/Mars Landers — Using AI for hazard detection, trajectory control, and safe landing
• NISR Satellite — Generates 85+ TB/day Earth observation data
• Netra Project — ISRO's collision avoidance for autonomous spacecraft
• Spacecraft Docking — 7 km/s precision docking autonomously (demonstrated by ISRO)

Datasets & Tools

• Apollo Landing Sites (Simulated) — Digital models for training autonomous landers
• Multimodal Earth Observation Data — Optical, SAR, drone sensors; standardized into common platform (e.g., VAS, BAN initiatives)
• Digital Twins — High-fidelity simulations of lunar/Martian terrain and spacecraft systems
• On-Ground Optimization Engines — Pre-process models before uploading to satellites (Skyserve methodology)

Policy & Funding

• AI-in-Space Seed Fund Scheme — ₹1 crore (~$120k USD) grants per startup; open to all AI-space applications
• Approved Disbursements — 11 startups already approved; 7 active, 4 under dispersement; 6 more planned

---

Broader Context & Significance

This session represents India's strategic pivot toward autonomous, AI-enabled space systems as a competitive advantage. Key implications:

• Sovereignty & Self-Reliance: Development of indigenous robotic systems, edge AI models, and autonomous missions reduces dependence on international partnerships for critical capabilities.
• Civil-Military Dual Use: Technologies (autonomous docking, collision avoidance, real-time GNC) apply to both civilian and defense satellite operations.
• Economic Opportunity: Space applications (agriculture, carbon credits, disaster management, climate monitoring) directly monetize satellite data for rural India.
• Workforce Development: Lowered barriers (startups, seed funding, multidisciplinary hiring) are attracting young talent to space sector.
• Global Convergence: Physics-based validation, explainability, and cautious integration align India's approach with international best practices (NASA, ESA) rather than rushing to full autonomy.

---

End of Summary