# HeMeR preprint reports faster inference in real-world robotic tests

_Published Saturday, November 14, 2026 at 7:06 PM EST · Science, AI · Latest · Tier 2 — Notable_

Researchers report that HeMeR reduced per-step inference time from 2.94 to 1.87 seconds compared with MemRL in real-world robotic deployments, with task success of 83.20%. The preprint has not been peer reviewed.

The framework tracks an agent's explicit knowledge, learned routines and current observations. It preserves cached model calculations for unchanged memory contents and updates only the calculations affected by memory changes, avoiding full recomputation as the agent makes successive decisions. The reported timing and success figures apply to the real-world setting evaluated by the researchers.

## Sources

- [cs.AI updates on arXiv.org](https://arxiv.org/abs/2611.16487)

---
Canonical: https://techandbusiness.org/newswire/bvgM9XYy0aushOzXKaXZJ1
Published: 2026-11-15T00:06:45.310Z
Story chronology: 2026-11-14T05:00:00.000Z
Retrieved: 2026-11-15T01:02:17.605Z
Publisher: Tech & Business (techandbusiness.org)