エピソード

  • The Robotics Brief — Robots, AI, and Autonomous Systems
    2026/09/02
    Dive into the cutting edge of robotic manipulation with Facet-0, a new foundation model tackling contact-rich tasks, and discover how SG-AMP is revolutionizing active perception and motion planning for agricultural robots. We then explore whether language models can reason about component lifecycles in complex agent systems with CordisBench, and examine the growing impact of verbal reinforcement learning. Finally, understand the critical role of mechanism design in achieving robust alignment and control for advanced robotic and AI systems.
    続きを読む 一部表示
    14 分
  • The Robotics Brief — Evolving AI Agents and World Models
    2026/08/29
    This episode dives into the cutting edge of AI agent evolution and advanced video modeling. Discover how agents gain persistent knowledge for skill evolution and even become automatic red-teamers, enhancing robustness.
    続きを読む 一部表示
    12 分
  • The Robotics Brief — AI Agents, Tools, and Sensing
    2026/08/22
    Unpack the latest in AI, from benchmarking LLM agents for recursive self-improvement in algorithmic design to inducing task models directly from computer-use traces. We also explore groundbreaking surgical robotics, examining joint visual-trajectory forecasting crucial for precise motion planning. Learn how MidTool enhances agentic tool use through mid-training data synthesis, alongside a fascinating comparison of radar technologies for non-invasive in-bedroom activity and sleep monitoring.
    続きを読む 一部表示
    16 分
  • The Robotics Brief — Robots Learn and Navigate 3D Worlds
    2026/07/26
    Unpack how robots are learning to manipulate with greater intelligence, from AXIS's community-driven data engine to strategic data collection that fosters compositional generalization. We then delve into critical advancements addressing memory bottlenecks in sequential embodied AI and equipping Vision-Language Models with robust 3D awareness. Finally, explore OpenForgeRL, a versatile platform for training agents across any environment, driving the next generation of adaptable robotic systems.
    続きを読む 一部表示
    15 分
  • The Robotics Brief — Robots, Safety, Scene Understanding, Causality, Memory
    2026/07/18
    Explore how advanced robotics are making strides in understanding and interacting with the world this week. We analyze RoboTTT's context scaling for smarter robot policies and confront the critical challenge of hidden physical dangers in robot actions, moving beyond mere text safety. Discover SceneBind's innovative approach to multimodal scene understanding across vision, audio, and language. Plus, we break down causal analysis for urban driving data and examine online neural space-time memory for dynamic novel view synthesis.
    続きを読む 一部表示
    15 分
  • The Robotics Brief — AI Agents and Multimodal Model Advancement
    2026/06/02
    Discover how researchers are making Multimodal LLMs fairer judges by mitigating perceptual bias and enhancing their continuous learning capabilities for new instructions. Explore groundbreaking methods for scalable robot data synthesis with compositional world models and a new predictive visual code for efficient video processing by MLLMs. Plus, delve into an innovative interactive environment simulating multi-stage clinical scenarios using electronic health records for robust AI agent training.
    続きを読む 一部表示
    12 分
  • The Robotics Brief — Generative Models for Reinforcement Learning
    2026/06/27
    Explore the cutting edge of generative AI as we dive into groundbreaking research transforming robotics and language models. Discover how DanceOPD and World Action Models are pushing the boundaries of continual imitation learning, alongside new approaches
    続きを読む 一部表示
    9 分
  • The Robotics Brief — Robots Learn, Adapt, and Understand
    2026/07/01
    Explore cutting-edge advancements in robot learning and manipulation this week. We dive into new world models like DVG-WM and AdaJEPA that boost efficiency and adaptability for complex tasks. Discover how robots are learning from freeform human preferences and even translating human video demonstrations into humanoid control with Human-as-Humanoid. Plus, uncover how metacognitive feedback in reinforcement learning can make LLMs express uncertainty more faithfully.
    続きを読む 一部表示
    14 分