Two five-second pauses made a simulated robot count work it had not done
Text
A controller mistook waiting for progress. Successes fell from 47 to 16 out of 50, then recovered to 45 when pauses were hidden from memory.
Asked to pick up and place a cube three times, a simulated robot stopped after two. Nanyang Technological University researchers found that two five-second pauses could make a memory-equipped controller treat waiting as work.
Counting-task successes fell from 47 to 16 out of 50. Keeping the pauses but excluding their frames and elapsed time from memory restored 45 successes.
Steadily paced demonstrations can teach a shortcut: elapsed time predicts progress. Accelerating memory timestamps also made the model finish early. The authors released code so other teams can examine the failure.
Event-based memory recovered some performance without retraining. But the November 10 preprint reports simulations, and its proposed fix relies on simulator signals. Real-robot reliability remains untested.
Source times: Paper, November 11, 2026, 02:47:57 Beijing / November 10, 1:47:57 PM EST; initial code release, November 10, 13:23:24 Beijing / November 10, 12:23:24 AM EST.
Share this article
What do you think of this story?
Anonymous feedback, up to 5 reactions per article.
Anonymous comments
Share your thoughts without signing in.
Discuss the technology and the story. Hate, politics and war, threats, abuse, personal accusations, sexual content, ads, links and personal information are not allowed.
No comments yet. Be the first to share.