IROS 2022 Keynote: James Kuffner -- The Future of Intelligent Machines

In this talk

1. 0:07

   Robots need to plan in real time

   Kuffner connects transportation, manufacturing, information technology, and AI to the rise of connected automated vehicles. Robots working outside factories face unknown objects and obstacles, so they need online perception and motion planning rather than fixed models and precomputed paths. High-dimensional search, uncertainty, and real-time safety constraints make this difficult.

2. 3:39

   Motion planning gains from computing power

   Kuffner describes how faster computers make it practical to search possible actions with forward simulators and heuristics. He and Steve LaValle developed rapidly-exploring random trees, including RRT-Connect, which grows trees from the start and goal to find collision-free paths. He says its practical balance between optimality and usable results helped it find broad use in robotics.

3. 8:38

   Combine learned models with physics

   Kuffner traces progress in robotics to on-demand data centers, connected fleets, and machine learning, including neural networks for vehicle perception. At Toyota, Advanced Drive brought hands-off highway driving to the second-generation Mirai and Lexus LS 500. He argues that learned models should complement physics and analytical models, supporting safety, explainability, and testing rare events in simulation.

4. 16:10

   Build systems that support people

   Kuffner says reliable automated driving depends on offboard software for training, simulation, developer tools, mapping, and data management, as well as onboard software. Woven City is planned as a living test site for mobility and robotics, with a digital twin for testing scenarios such as charging-station placement and delivery times. He frames the goal as amplifying human abilities, including helping people age in place.