Ep#100: 100 Episodes of Embodied AI

In this talk

1. 3:04

   How RoboPapers began

   The hosts trace the show to Michael’s idea for a podcast that examines robotics papers in technical depth. They began without a production plan, and say it took about 10 episodes to settle basic recording and captioning workflows. RJ, who has worked in robotics and self-driving for a decade, joins as the fourth co-host to help cover areas such as navigation.

2. 8:45

   A curious, community-focused format

   The hosts say the show helps them learn from researchers and ask questions that papers often leave unanswered, including what failed and why particular choices were made. They select papers out of individual curiosity, while acknowledging that this has left topics such as navigation and human-robot interaction underrepresented. Their aim is to make technical work accessible to early-career researchers and other interested listeners.

3. 38:17

   Trends and open questions in robotics

   The hosts describe more work on whole-body control, sim-to-real, world models, on-robot reinforcement learning, and larger open-source datasets. Chris and RJ remain skeptical of sim-to-real without strong real-to-sim pipelines. They also question whether large-scale data and compute alone explain new robot capabilities, and argue that diverse research approaches and better ways to measure progress matter.

4. 51:25

   What 100 episodes reveal

   The hosts report 100 episodes, 153 guests, more than 50 labs and companies across nine countries, and close to 100 hours of conversation. Their informal analysis suggests coverage is concentrated in the United States, with little representation from Japan and Korea and fewer episodes on navigation. They caution that AI-labeled themes and a 100-episode sample are not rigorous statistics.

5. 1:17:08

   Plans for the next 100

   The hosts want to cover more theory, hardware, safety, deployment, HRI, soft robotics, planning, and research from underrepresented regions. They are considering publishing on Xiaohongshu and Bilibili, recording some in-person episodes, holding a community event, and starting a Discord server. They emphasize keeping the show technically grounded, impartial, and supportive of open research.