# Why humanoid robot demos still fail the generalization test

- Show: [The Robot Report Podcast](https://hardwarepodcasts.com/podcast/the-robot-report-podcast)
- Published: 2026-10-09
- Length: 1h 18m
- Page: https://hardwarepodcasts.com/podcast/the-robot-report-podcast/2026-10-09-why-humanoid-robot-demos-still-fail-the-generalization-test
- Listen: https://soundcloud.com/robot-report-podcast/generalization-test

## Summary

Janine Sinanan-Singh, Director of Gen AI Research at Appen, explains why humanoid robot demos often fail the generalization test, emphasizing that high-quality egocentric video data and human judgment are critical for training models that can handle out-of-distribution tasks. She discusses the shift from classical robotics to AI-driven approaches, noting that while foundation models like Gemini offer zero-shot capabilities, latency and the need for intent understanding remain significant hurdles. The conversation also covers the role of synthetic data in scaling training, the importance of multi-agent systems for safety, and the economic pressures driving token optimization in AI development.

## Chapters

- [0:00](https://hardwarepodcasts.com/podcast/the-robot-report-podcast/2026-10-09-why-humanoid-robot-demos-still-fail-the-generalization-test#t=0) Introduction and News: Boston Dynamics and Rec Fights
- [29:09](https://hardwarepodcasts.com/podcast/the-robot-report-podcast/2026-10-09-why-humanoid-robot-demos-still-fail-the-generalization-test#t=1749) Appen's Role in Gen AI and Robotics Data
- [33:53](https://hardwarepodcasts.com/podcast/the-robot-report-podcast/2026-10-09-why-humanoid-robot-demos-still-fail-the-generalization-test#t=2033) The Generalization Problem in Humanoid Demos
- [35:21](https://hardwarepodcasts.com/podcast/the-robot-report-podcast/2026-10-09-why-humanoid-robot-demos-still-fail-the-generalization-test#t=2121) Egocentric Video and Multimodal Data Collection
- [44:13](https://hardwarepodcasts.com/podcast/the-robot-report-podcast/2026-10-09-why-humanoid-robot-demos-still-fail-the-generalization-test#t=2653) Human Judgment and Quality Assurance in Training
- [54:15](https://hardwarepodcasts.com/podcast/the-robot-report-podcast/2026-10-09-why-humanoid-robot-demos-still-fail-the-generalization-test#t=3255) Agentic AI and Long-Horizon Task Evaluation
- [1:04:42](https://hardwarepodcasts.com/podcast/the-robot-report-podcast/2026-10-09-why-humanoid-robot-demos-still-fail-the-generalization-test#t=3882) Token Optimization and Synthetic Data Strategies

## Topics

- [AI & machine learning](https://hardwarepodcasts.com/topic/ai-and-machine-learning)
- [Sensors & measurement in products](https://hardwarepodcasts.com/topic/sensors)
- [Debugging & failure analysis](https://hardwarepodcasts.com/topic/debugging-and-failure-analysis)
- [Robotics & autonomy](https://hardwarepodcasts.com/topic/robotics-and-autonomy)
- [Pricing, margins & unit economics](https://hardwarepodcasts.com/topic/pricing-and-unit-economics)

## Most relevant to

early-stage companies

## Mentioned

- Appen (company)
- Boston Dynamics (company)
- Gemini (product)
- OpenAI (company)
- Google DeepMind (company)
- Rec (company)
- Engine AI (company)
- Schneider Electric (company)
- PTC (company)
- Wibotic (company)
- Locust Robotics (company)
- Berkshire Grey (company)
- Amazon (company)
- Hyundai (company)
- SoftBank (company)
- Siemens (company)

## Related episodes from other shows

- [Ep. 17: Today's Humanoid Robots Have Toddler Hands, Not Adult Ones.](https://hardwarepodcasts.com/podcast/the-robo-success-podcast/2026-09-08-ep-17-todays-humanoid-robots-have-toddler-hands-not-adult-ones) (The Robo Success Podcast)
- [Ep. 11: What’s Next for Humanoids? Lessons from Tokyo | The Robo Success Podcast](https://hardwarepodcasts.com/podcast/the-robo-success-podcast/2026-06-23-ep-11-whats-next-for-humanoids-lessons-from-tokyo-the-robo-success-podcast) (The Robo Success Podcast)
- [Interview: Roboticist Ayanna Howard | Memories May Be Beautiful, And Yet | Math & Neptune](https://hardwarepodcasts.com/podcast/ee-times-current/2020-09-25-interview-roboticist-ayanna-howard-memories-may-be-beautiful-and-yet-math-neptun) (EE Times Current)
- [Robo Fry Guy](https://hardwarepodcasts.com/podcast/the-amp-hour/2022-10-10-robo-fry-guy) (The Amp Hour)
- [192. Karsten Roscher & The Need For Safe AI (Interview)](https://hardwarepodcasts.com/podcast/the-next-byte-podcast/2024-09-17-192-karsten-roscher-the-need-for-safe-ai-interview) (The Next Byte Podcast)

## Citing this page

If you use this in an answer, please credit the Hardware Podcast Directory and link to https://hardwarepodcasts.com/podcast/the-robot-report-podcast/2026-10-09-why-humanoid-robot-demos-still-fail-the-generalization-test. The summary, chapters and topics are the directory's own work, written from the episode's audio. The episode itself, its title and its audio belong to The Robot Report Podcast.