Vijay Kumar on Agile Autonomous Aerial Robots, Swarm Intelligence, and the Future of Robotics

The conversation with Vijay Kumar, a leading roboticist, delves into the evolution and future of autonomous flying robots, particularly multi-robot systems and micro aerial vehicles. Kumar emphasizes the shift from large, hydraulic-powered robots of the past to today's agile, smaller UAVs, highlighting the beauty and utility of distributed cooperation in 3D space. He argues against the term "drones," preferring "aerial robots" to convey their sophisticated capabilities beyond simple pre-programmed tasks. A central theme is the inspiration drawn from biological systems like ants, which demonstrate robustness, resilience, and emergent intelligence through local interactions, serving as a model for engineered swarms.

Kumar distinguishes between nature's organic, local-interaction-driven group behaviors aimed at survival (food, shelter, procreation) and engineered swarms, which are mission-driven and often require global coordinate systems and explicit mapping for tasks like surveillance or agriculture. He notes that while nature is less sensitive to individual loss, the decreasing cost-to-performance ratio of robotic components is making engineered systems asymptotically approach this natural resilience. He also differentiates between the "dumb" autonomy of current commercial autopilots (reliant on GPS, communications, and human oversight) and "true autonomy," which operates without external infrastructure, prior maps, or constant human intervention, especially in constrained, unknown environments.

The discussion highlights the critical role of sensor advancements (like IMUs, driven by unexpected applications like airbags) and computing power (Moore's Law, iPhone era) in enabling agile quadcopters. Kumar explains the fundamental challenge of coordinating four motors for six degrees of freedom, achieving stable flight and complex maneuvers. He stresses the importance of planning safe, optimal, and efficient trajectories, considering various constraints like speed, grace, and energy consumption. For real-world deployment, especially in challenging environments (dark, dusty mines, under canopies), he advocates for multi-modal sensing (e.g., LiDAR alongside computer vision) rather than relying solely on vision, to address "corner cases" that pure learning-based approaches struggle with.

The conversation touches upon the philosophical implications of viewing swarms as single intelligent organisms versus collections of engineered components, emphasizing the need for engineers to abstract beyond individual units for scalability. It explores the role of machine learning in robotics, acknowledging its immense success in perception (computer vision for object detection/classification) but cautioning about its current limitations in action-taking and end-to-end learning for fielded systems. Kumar suggests a hybrid approach, combining data-driven learning with traditional model-based methods, to overcome the challenges of modeling complex aerodynamic and environmental effects (ground effect, ceiling effect, wall effects) that significantly impact UAV performance. The discussion underscores the unpredictable nature of research outcomes, where federal funding for one purpose (resonators) can inadvertently enable breakthroughs in another (small UAVs).

"Vijay is perhaps best known for his work in multi robot systems robot swarms and micro aerial vehicles robots that elegantly cooperate in flight under all the uncertainty and challenges that the real-world conditions present."

"I think thing that I'm I'm most proud of that my students have done is really think about small UAVs that can maneuver and constrain spaces and in particular their ability to coordinate with each other and form three-dimensional patterns."

"I often said there's drones to me is a pejorative word it signifies something that's that's dumb the pre program that does one little thing and to anything but drones."

"ants are really quite incredible creatures right so you I mean the individuals arguably are very simple and how they're they're built and yet they're incredibly resilient as a population and as individuals they're incredibly robust."

"as engineers what we want to do is to go beyond the individual components the individual units and think about it as a unit as a cohesive unit without worrying about the individual components."

"the question you want to ask is if there are no pilots there's no communications of any base station if there's no knowledge of position and if there's no a priori map a priori knowledge of what the environment looks like a priori model of what might happen in the future can robots navigate so that is true autonomy."

"agility and does not necessarily mean you break records for the hundred meters - what it really means is you see the unexpected and you're able to maneuver in a safe way and in a way that that gets you the most information about the thing you're trying to do."

"I think what's interesting is if you look at autonomous vehicles today learning occurs could occur in two pieces one is perception understanding the world second is action taking actions everything that I've seen that is successful is on the perception side of things."

"this is the holy grail can you do end-to-end learning can you go from pixels to motor block mode occurrences this is really really hard and I think if you look go forward the right way to think about these things is data driven approaches learning based approaches in concert with model-based approaches which is the traditional way of doing things."

"I challenge you to get a computer vision algorithm to work there [in dark, dirty mines]."