# \#kepler

**URL:** https://forum.menlo.ai/tag/kepler/31.md

[Latest](https://forum.menlo.ai/latest.md) · [Categories](https://forum.menlo.ai/categories.md) · [Tags](https://forum.menlo.ai/tags.md)

---

## [Why VLAs are so viewpoint brittle?](https://forum.menlo.ai/t/why-vlas-are-so-viewpoint-brittle/66)

<div class="topic-metadata">

**Author:** [@Ishneet\_Singh](https://forum.menlo.ai/u/Ishneet_Singh)\
**Replies:** 4\
**Last updated:** [September 29, 2026, 3:47am UTC](https://forum.menlo.ai/t/why-vlas-are-so-viewpoint-brittle/66 "2026-09-29T03:47:11Z")

</div>

TL;DR: a tripod nudge can push the policy much further out of distribution than a completely new object layout. Fixing this likely requires multi-view training, viewpoint augmentation, or representations explicitly encou…

---

## [A new arm is free. A new gripper is not](https://forum.menlo.ai/t/a-new-arm-is-free-a-new-gripper-is-not/68)

<div class="topic-metadata">

**Author:** [@Ishneet\_Singh](https://forum.menlo.ai/u/Ishneet_Singh)\
**Replies:** 1\
**Last updated:** [September 22, 2026, 1:47am UTC](https://forum.menlo.ai/t/a-new-arm-is-free-a-new-gripper-is-not/68 "2026-09-22T01:47:07Z")

</div>

TL;DR: I ran a frozen \`pi05\_libero\` on arms it was never trained on. UR5e and the I2RT YAM both go 20/20, same as the Panda it was fine-tuned on, as long as the wrist camera sits in the same place relative to the fingers…

---

## [This week's experiment update: Kepler v0.2](https://forum.menlo.ai/t/this-weeks-experiment-update-kepler-v0-2/63)

<div class="topic-metadata">

**Author:** [@linus](https://forum.menlo.ai/u/linus)\
**Replies:** 0\
**Last updated:** [August 4, 2026, 10:06am UTC](https://forum.menlo.ai/t/this-weeks-experiment-update-kepler-v0-2/63 "2026-08-04T10:06:27Z")

</div>

One of our goals with Menlo is to share our research process, including experiments that don’t go as expected. This week, we ran follow-up experiments on Kepler v0.1 to understand why our pretrained encoder wasn’t impro…

---

## [Why Didn't a Better Encoder Produce a Better Robot Policy?](https://forum.menlo.ai/t/why-didnt-a-better-encoder-produce-a-better-robot-policy/62)

<div class="topic-metadata">

**Author:** [@linus](https://forum.menlo.ai/u/linus)\
**Replies:** 0\
**Last updated:** [August 4, 2026, 6:04am UTC](https://forum.menlo.ai/t/why-didnt-a-better-encoder-produce-a-better-robot-policy/62 "2026-08-04T06:04:09Z")

</div>

Last week we shared that our pretrained encoder learned good representations but didn’t improve downstream policy performance. Our hypothesis was that the issue was the training objective rather than the architecture, so…

---

## [This week's Kepler v0.2 experiment updates](https://forum.menlo.ai/t/this-weeks-kepler-v0-2-experiment-updates/56)

<div class="topic-metadata">

**Author:** [@linus](https://forum.menlo.ai/u/linus)\
**Replies:** 0\
**Last updated:** [July 27, 2026, 6:02am UTC](https://forum.menlo.ai/t/this-weeks-kepler-v0-2-experiment-updates/56 "2026-07-27T06:02:09Z")

</div>

One of our goals with Menlo is to share our research process, including experiments that don’t go as expected. This week, we ran follow-up experiments on Kepler v0.1 to understand why our pretrained encoder wasn’t impro…