NVIDIA GEAR Lab launches RoboTTT to scale robotic memory to 8000 timesteps

Users are excited about NVIDIA's RoboTTT for scaling robot policies to 8K timesteps because it shows impressive technical results with zero saturation on the performance curve.

Based on 2 visible X reactions from 2 accounts; directional sample.

Useful Links

RoboTTT: Context Scaling for Robot Policies(nvidia.com)viaJim Fan

Digg Deeper

Ask a question below.

Published answers will appear here.

Commentary on X

YJ

Yunfan Jiang@YunfanJiang

@scott_e_reed Thank you, Scott! Really looking forward to seeing what you build next, too. 🙌

8:02 AM · Jul 15, 2026

Reply

GT

GTechne@ai4urbanlife

@DrJimFan slap some TTT on it 😆 but seriously, zero saturation on that curve is wild

8:32 AM · Jul 15, 2026

Reply

JF

Jim Fan@DrJimFan

Blog and paper: https://research.nvidia.com/labs/gear/robottt/ Presented to you from NVIDIA GEAR Lab. Please check out Yunfan's technical deep dive!

8:25 AM · Jul 15, 2026

Reply

SR

Scott Reed@scott_e_reed

Awesome project! I think TTT is great for robot learning, and should be agnostic to VLA vs video backbone.

7:49 AM · Jul 15, 2026

Reply

The method maintains constant inference latency using test-time training.

SR

Scott Reed@scott_e_reed

Awesome project! I think TTT is great for robot learning, and should be agnostic to VLA vs video backbone.

7:49 AM · Jul 15, 2026

Reply

9 saves

1 reposts

Sentiment

Positive100%0%Negative

Positive Read

Users are excited about NVIDIA's RoboTTT for scaling robot policies to 8K timesteps because it shows impressive technical results with zero saturation on the performance curve.

Based on 2 visible X reactions from 2 accounts; directional sample.

Useful Links

RoboTTT: Context Scaling for Robot Policies(nvidia.com)viaJim Fan

Digg Deeper

Ask a question below.

Published answers will appear here.