DAY 2 | FREE TRACK
Robots, Autonomous Systems, and AI that Acts in the Real World
9:45 - 10:00
Chairperson’s opening remark
Chairpersons welcome and opening remarks
10:00 - 10:40
Panel: The Physical AI Stack - From Edge Devices to Autonomous Action
A macro view of how physical AI systems are structured:
- Edge AI (STBs, sensors, smart cameras)
- Multimodal perception & sensor fusion
- Foundation models for robotics
- Orchestration layers
- Governance & safety
10:45 - 11:15
Presentation: Physical AI in 2026 - from pilots to scaled fleets
What is genuinely new (foundation models, simulation at scale, edge compute), and what still breaks in production.
11:20 - 11:40
Presentation: Physical AI for Industrial Maintenance and World‑Model‑Driven Decision Intelligence
Industrial assets don’t fail in isolation: operating regimes shift, components influence one another, and data is often sparse or incomplete. This talk will introduce our Physical‑AI approach that combines an asset‑agnostic simulation twin, LLM agents, and a learned world model to deliver prescriptive maintenance at scale. The twin lets us generate thousands of trajectories to learn state‑transition dynamics and uncertainty and train a world model that can be used for what‑if analysis, counterfactual diagnostics, and policy optimization across reliability, safety, and cost. LLMs accelerate configuration and explainability, extracting parameters from manuals and work orders, and translating engineer intent into experiments while the world model does the reasoning: forecasting futures, testing interventions, and quantifying trade‑offs.
We’ll explore how this Physical‑AI framework is being developed and refined, and outline the emerging pathways toward enterprise integration—including the research directions, design choices, and practical considerations shaping its evolution.
11:40 - 12:10
Networking Break
12:10 - 12:40
Presentation: Accelerating Autonomous Mobility with NVIDIA DRIVE
Exploring how NVIDIA DRIVE enables autonomous vehicle development at scale through advanced AI models, high-performance compute, and simulation. This session will cover perception, decision-making, and continuous learning pipelines that power safe, efficient automated driving systems and fleet-level optimisation.
12:45 - 1:15
Presentation: SDVs & Physical AI
Vehicles aren’t “just cars” — they’re mobile AI platforms that perceive city infrastructure, interact with people and machines, and make real-time decisions. This session covers:
Perception stacks in mobility vs robotics
Predictive models for unstructured urban interaction
Sensor fusion between vehicles and infrastructure
Safety, simulation, and regulation parallels with robots
1:15 - 2:05
Lunch Break
2:05 - 2:45
Panel: Robots, Autonomous Systems & the New Embodied Intelligence Stack
Join us as we take a deep dive into robotics – but as part of the bigger ecosystem.
- Robot foundation models
- Vision-language-action systems
- Generalisation challenges
- Build vs buy decisions
- Where humanoids fit (and don’t)
2:50 - 3:20
Presentation: The Economics of Physical AI – Workforce, ROI & Transformation
- Labour displacement vs augmentation
- Tele-ops & human-in-the-loop models
- Measuring ROI properly
- Enterprise adoption timelines
- What becomes autonomous first?
3:25 - 3:55
Presentation: Scaling Physical AI with WFMs and VLA Models
The path to general-purpose robotics – Physical AI – has long been obstructed by the “Data Wall.” Traditional imitation learning requires thousands of hours of human demonstrations, making it nearly impossible to scale robot skills across diverse environments and complex tasks. This talk explores a new paradigm in robotic training: leveraging World Foundation Models (WFMs) to generate massive-scale synthetic trajectory data.
We will examine the end-to-end pipeline for scaling Physical AI, beginning with the use of WFMs to “imagine” and simulate realistic future world states. We will discuss how these high-fidelity synthetic scenarios are converted into actionable data to train Vision-Language-Action (VLA) models, enabling robots to understand and execute natural language instructions in unseen settings. Finally, we will cover the critical role of policy evaluation and the “real-to-real” workflow, demonstrating how these technologies collectively reduce development timelines from months to hours while significantly increasing the generalization of robotic agents.
