Ai

The integration of AI into robotics has transformed how robots and autonomous systems perceive, learn, and act across industries: from intelligent bin-picking to collaborative tasks in modern factories. Now, with the rise of Generative AI, we’re witnessing a fundamental shift in how robotics systems are built, trained, and deployed.

This talk will discuss how to develop intelligent robotic systems using both traditional AI approaches and the latest advancements in robotics foundation models. Learn how to design end-to-end workflows that incorporate deep learning, reinforcement learning, transformer-based vision-language-action (VLA) models all within a single, simulation-driven platform.

Highlights:
• Design and deploy AI-powered bin-picking and motion planning systems with reduced human supervision.
• Automate data labeling and training for object detection and pose estimation.
• Object detection with zero-shot text-conditioned models.
• Segmenting objects across images and videos using vision foundation models

As robotics shifts from task automation to intelligent autonomy, the real challenge isn’t just building robots that move, it’s designing systems that learn, adapt, and earn trust at scale. John Black, CTO at Brain Corp, will unpack what it takes to build and deploy robot fleets that combine reliability with intelligence.

Drawing from Brain Corp’s experience powering over 40,000 robots worldwide, Black will explore how new approaches in perception, data integration, and continuous learning are redefining autonomy in real-world environments like retail, logistics, and public spaces. He’ll share how teams can bridge the gap between simulation and deployment, optimize for safety and privacy by design, and architect systems that evolve responsibly over time.

Attendees will gain a behind-the-scenes view of how to turn complex, dynamic environments into structured intelligence, and practical lessons on scaling robot software, managing edge-to-cloud data flows, and maintaining trust with users and regulators alike.

As your robotics software development organization scales, the developer experience grows both more important and harder to get right–and an inefficient build system can cost you in slow iteration cycles, developer frustration, and wasted resources.

We will share insights from the Robotics and AI Institute’s DevExp and build system overhaul: a project with a scope spanning from repository organization (mono-repo or multi-repo?) to polyglot build systems with caching and remote build execution, ROS integration, management of (conflicting!) dependencies, all the way to runtime tools, deployment, and migration–for an organization consisting of many different robotics research and engineering teams with diverse needs.

We’ll discuss our process, the choices we made and why, as well as other options we considered, bespoke tools we had to build, and retrospective lessons and future plans. Come to learn how we’re saving 164 CPU-hours per day, how we make Bazel and ROS work together, or how close we came to total failure before delighting users.

Robots don’t get smarter by accident; they improve through a deliberate data flywheel. This panel explores how real-world robot data is captured, curated, and fed back into perception, autonomy, and system performance. Drawing on lessons from field-deployed robots, we’ll explore how to design the data flywheel from day one, turning pilots into scalable, field-ready platforms that continuously learn and improve in production.

We believe that ROS and our other open-source projects at Open Robotics are essential to the next generation of robotics. This principle drives governance, development, and collaboration within our Open Source Robotics Alliance (OSRA), and we continue to see an expanding global community and ecosystem as the industry increasingly shares this view.

This session will include an overview of the current state of open source in robotics and AI and a discussion about our strategic priorities and plans under the OSRA:

Embrace the AI revolution – be the premier open source platform for developing AI-driven robotic systems

Expand accessibility and ease of use – make our projects much easier to install and use across all major platforms

Adopt modern tools from industry – empower professional developers with enhanced support for modern tools

Implementation/Focus Areas:

The Physical AI Special Interest Group (SIG) – key achievements and immediate goals, prominent contributors

Incorporation of principles into ROS, ros-controls, Gazebo, Open-RMF, and Infrastructure governance, roadmap features, and development

Collaboration with key community stakeholders to drive adoption and knowledge sharing

Leaders from Path Robotics, Universal Robots, and PickNik Robotics will explore how artificial intelligence is making today’s robots more capable, adaptable, and easier to deploy. The discussion will take a practical, engineering-focused look at what AI can and can’t do in real-world production environments, from vision-guided welding and collaborative manipulation to motion planning and autonomy.

The panel will examine the current state of AI in robotics: how leading companies are integrating machine learning, foundation models, and advanced perception into commercial systems; the infrastructure and data challenges involved; and the lessons learned from deploying AI-enabled robots at scale.

Robots are evolving from programmed ROS applications to embodied AI (LLMs running on physical robotics hardware) solving missions. New robotics AI tools are evolving such as VLMs (Vision Language Models also called multimodal LLMs), VLAs (Vision Language Action models), and MCP (Model Context Protocol) servers. Robots now stream sensors like RealSense depth cameras into these AI tools allowing the AI to figure out on its own how to move its wheels or legs to achieve a goal such as following a person. During this session, we will build an embodied AI experience together live on stage! What could go wrong?!

Join us for a dynamic showcase featuring startups from the Physical AI Fellowship, powered by AWS, NVIDIA, and MassRobotics. Over the past eight weeks, these companies worked closely with the AWS GenAI Innovation Center technical team to validate, test, and refine their robotics and AI technologies. This session will highlight some of the most promising innovations emerging at the intersection of physical AI, autonomy, and real-world deployment. Attendees will gain insight into the startups’ technologies, the challenges they tackled, and the lessons learned through deep collaboration with industry leaders. Expect fast-paced presentations, cutting-edge demos, and a firsthand look at the future of intelligent machines.

What if you could control a robot using natural language? In this session, we demonstrate how to enable natural language as a control layer for Physical AI, using AWS Strands with the Boston Dynamics Spot robot as a real-world example. We’ll walk through how user intent expressed in plain language is translated into structured plans, tool calls, and robot actions—bridging large language models with perception, navigation, and actuation.