Robot Talk · 19 May 2023 · 33m · Episode 49
Nick Hawes, Professor of AI and Robotics at the University of Oxford and director of the Oxford Robotics Institute, discusses his GOALS group's work on goal-oriented decision-making for autonomous systems under uncertainty. He details the challenges of long-term autonomy, citing a month-long continuous run of a mobile robot in the Strands project, and explains how hardware degradation and environmental changes complicate sustained operation. Hawes also describes the Hudson project, where an autonomous underwater vehicle (Nico Sub) harvests data from a sensor network in Loch Ness, optimizing routes based on epistemic and aleatoric uncertainty.
Where this conversation describes AI being used to build or deliver a product. Part of the AI in practice tracker.
Engineering & simulation
The team used machine learning to analyze statistics of how often doors opened at different times of day, creating models of environmental dynamics that allowed the robot to adapt its behavior over time.
“So we used machine only to kind of pick up those sort of statistics to give us models of the way the robot behaved and that's the environment behaved when the robot was there.”
Speaker 2 · Hear it at 7:31
Engineering & simulationreinforcement learning
The team used reinforcement learning to learn mappings from data for robot policies, though they noted it is currently more experimental than their model-based approaches for long-term guaranteed behavior.
“But you can use something like reinforcement learning to learn those mappings from data as well. So we use those techniques as well for the robots. … the reinforcement learning for robots for us is a bit more experimental at this thing.”
Speaker 2 · Hear it at 24:49
early-stage companies
Other shows on the same ground, closest first.
Marc Raibert, founder of Boston Dynamics and director of the AI Institute, discusses his transition from leading Boston Dynamics for 30 years to establishing a new long-term R&D organization backed by Hyundai. He outlines the AI Institute's four pillars—cognitive AI, athletic AI, organic hardware, and ethics—and describes ambitious goals like robots learning tasks by observation and performing complex mobile manipulation. Raibert also addresses the current state of humanoid robotics, the challenges of battery life and dexterity, and his vision for a future where robots assist rather than replace human labor.
Nic Radford, founder and CEO of Nauticus Robotics, discusses the development of the Aquanaut, a hybrid autonomous underwater vehicle (AUV) and remotely operated vehicle (ROV) that combines the mobility of a torpedo with the manipulation capabilities of a box-shaped ROV. Leveraging his background at NASA with the Robonaut 2, Radford explains how Nauticus applies space-grade delay-tolerant networking and AI to overcome the extreme pressure and low-bandwidth challenges of the deep ocean. The company operates on a Robots-as-a-Service (RaaS) model, deploying the Aquanaut alongside their autonomous surface vessel, Hydronaut, to perform inspection and maintenance tasks for offshore wind and oil and gas industries without the high costs of traditional tethered spreads. Radford highlights the difficulty of recruiting hardware talent and the lack of venture capital for industrial deep tech, emphasizing that their proprietary electric actuation and software-defined autonomy create a significant competitive moat.
Michael Patrick Perry, VP of Business Development at Boston Dynamics, discusses the company's acquisition by Hyundai Motor Group for $880 million and its commercialization of the Spot quadruped robot, which has seen 400 deployments. Dr. Timothy Chung, program manager for the DARPA Subterranean Challenge, details the lessons learned from the virtual Cave Circuit competition, highlighting the importance of heterogeneous robot teams and autonomy in GPS-denied environments.
Elecia White discusses her recent contract work on autonomous off-road trucks, where she implemented reinforcement learning for a safety system and debugged a critical color-space issue in TensorFlow. She also details her preparation for the second cohort of her Making Embedded Systems course, focusing on firmware update security and IoT management. Christopher White shares his experience building a radar data collection system using an NVIDIA Xavier and Python GUIs, and reflects on the importance of practice and exercise.
Mo ElShenawy, Senior VP of Engineering at Cruise (a GM subsidiary), discusses the engineering challenges of developing fully autonomous vehicles (L4/L5). He explains why Cruise targets San Francisco's complex environment to accelerate learning and details the hardware stack, including LiDAR, radar, and cameras, fused for perception and prediction. ElShenawy emphasizes that because state-of-the-art literature cannot yet solve L4/L5 driving, Cruise is inventing new algorithms and building a massive simulation and data infrastructure to ensure safety and scale.
The summary, chapters and topics are generated from the episode’s audio by this directory, not written by Robot Talk. The episode itself belongs to its publisher.