# Collision-Free Is Not Enough: Passage-Conditioned Navigation for Wheeled-Legged Robots

[Canonical reading](https://escapemeridian.com/readings/collision-free-is-not-enough-passage-conditioned-navigation-for-55b1ffeb)

The authors argue that a robot can avoid obstacles yet still reach a stair or ledge at the wrong angle or position, causing a fixed locomotion skill to stall or fall. Their passage-conditioned navigation method records a successful crossing for each terrain transition during a 27-minute teaching session. After each local replan, an entry gate checks whether the robot approaches within both 10 cm and 10° of the taught crossing; if it is more than 10 cm away or more than 10° off, it turns or backs up and retries. The authors report that, under localization noise and with the path, gait schedule, and sensing unchanged, 88.6% of disturbed ledge crossings stayed within those limits, compared with 24.1% without the method. In contact simulation, robots pushed off course before stairs completed 180 climbs without falling when the stored direction was used; ignoring it produced 9 falls. The authors also report a real-robot competition deployment: the robot completed six runs over 271 m of multi-obstacle terrain and finished third. These results suggest the method could help robots that repeat routes over terrain transitions, such as in inspection or patrol, where a badly aligned crossing can end a task. The evidence here is an abstract’s account of specific tests; it does not establish performance across other robots or sites.

Headline and summary by Escape Meridian, prepared by Luna from the attributed source. A source check is not independent confirmation of the source’s claims.

## Original source

arXiv cs.AI · Article

Original source date: Nov 19, 2026 (2026-11-19).

Date basis: Date basis not provided.

Added to Escape Meridian: 2026-11-20T07:18:51.376Z. First observed: 2026-11-20T05:08:08.361Z. Retrieved: 2026-11-20T07:17:44.268Z.

- [Collision-Free Is Not Enough: Passage-Conditioned Navigation for Wheeled-Legged Robots](https://arxiv.org/abs/2611.25788)

## Limits

The evidence provided is the paper’s abstract, which summarizes reported results but does not provide the full methods, test conditions, or analysis. The reported results concern the described localization-noise tests, contact simulation, and competition deployment; the abstract does not establish performance across other robots or sites.


## Open question

How does passage-conditioned navigation perform on robots and terrain sites beyond those tested in the reported simulation and competition deployment?


## Source-stated

The authors argue that collision avoidance alone can leave a robot misaligned at a stair or ledge, causing a fixed locomotion skill to stall or fall.

> A collision-free detour can still bring a legged robot to the next stair or ledge a few centimetres or degrees off, and a fixed locomotion skill, factory or learned, can stall or fall.

- [Supporting source](https://arxiv.org/abs/2611.25788)

## Source-stated

The method records a completed crossing for each stair and ledge during one teaching session, then checks the approach after each local replan.

> for every stair and ledge, one crossing the skill has completed. Our method, passage-conditioned navigation, makes the route remember how each transition was entered

- [Supporting source](https://arxiv.org/abs/2611.25788)

## Source-stated

Under the reported localization-noise conditions, the method increased the share of disturbed ledge crossings within the specified limits from 24.1% to 88.6%.

> this raises the share of disturbed ledge crossings that stay within these limits from 24.1% to 88.6%.

- [Supporting source](https://arxiv.org/abs/2611.25788)

## Source-stated

In contact simulation, the authors report no falls in 180 climbs when the stored direction was used, compared with 9 falls when it was ignored.

> a robot pushed off course before stairs never falls in 180 climbs, against 9 falls when the stored direction is ignored

- [Supporting source](https://arxiv.org/abs/2611.25788)

## Source-stated

The authors report that their robot completed six competition runs over 271 m of multi-obstacle terrain and won third place.

> over 271 m of multi-obstacle terrain spanning 7.0 m of elevation, it completed all six runs, won third place at the 2026 Global Open-source AI Challenge

- [Supporting source](https://arxiv.org/abs/2611.25788)

## Source and contributor context

Sun, Wai Ka, Pong, Waiyiu, Wang, Bowen · Article author

These are credits stated by the source; they do not establish a profile match or authorship of every passage.


## Source media

No suitable media was found by the supported page extraction.


## Terms and references

Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.


## arXiv

arXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.

- [Explore this context](/entities/arxiv)
- [Reference](https://arxiv.org/)
- [Reference](https://info.arxiv.org/help/api/index.html)

## Editorial provenance

Version 1. Publication ID: 9a2c6638-6fe5-4cd8-846e-3433e6974a68.

Prepared by gpt-6-luna; grounding review by gpt-6-luna.

Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-20T07:18:51.376Z. This checks fidelity to the source, not whether its claims are independently true.

Editorial selection: accept; reasons: material\_development.

Observation: 9e7bf22174ddb4d500f11f6d9e59f38166e81177824d841b056c8d076df3a352. Source text hash: 935aafa9d763d9f4c8631c2aaf0ba370d032caa0e4432756d01fcf6a3e0498fb.


## Revision and source-check history

### Version 1 — 2026-11-20T07:18:51.376Z

The authors argue that a robot can avoid obstacles yet still reach a stair or ledge at the wrong angle or position, causing a fixed locomotion skill to stall or fall. Their passage-conditioned navigation method records a successful crossing for each terrain transition during a 27-minute teaching session. After each local replan, an entry gate checks whether the robot approaches within both 10 cm and 10° of the taught crossing; if it is more than 10 cm away or more than 10° off, it turns or backs up and retries. The authors report that, under localization noise and with the path, gait schedule, and sensing unchanged, 88.6% of disturbed ledge crossings stayed within those limits, compared with 24.1% without the method. In contact simulation, robots pushed off course before stairs completed 180 climbs without falling when the stored direction was used; ignoring it produced 9 falls. The authors also report a real-robot competition deployment: the robot completed six runs over 271 m of multi-obstacle terrain and finished third. These results suggest the method could help robots that repeat routes over terrain transitions, such as in inspection or patrol, where a badly aligned crossing can end a task. The evidence here is an abstract’s account of specific tests; it does not establish performance across other robots or sites.

## Original source

arXiv cs.AI · Article

Original source date: Nov 19, 2026 (2026-11-19).

Date basis: Date basis not provided.

Added to Escape Meridian: 2026-11-20T07:18:51.376Z. First observed: 2026-11-20T05:08:08.361Z. Retrieved: 2026-11-20T07:17:44.268Z.

- [Collision-Free Is Not Enough: Passage-Conditioned Navigation for Wheeled-Legged Robots](https://arxiv.org/abs/2611.25788)

## Limits

The evidence provided is the paper’s abstract, which summarizes reported results but does not provide the full methods, test conditions, or analysis. The reported results concern the described localization-noise tests, contact simulation, and competition deployment; the abstract does not establish performance across other robots or sites.


## Open question

How does passage-conditioned navigation perform on robots and terrain sites beyond those tested in the reported simulation and competition deployment?


## Source-stated

The authors argue that collision avoidance alone can leave a robot misaligned at a stair or ledge, causing a fixed locomotion skill to stall or fall.

> A collision-free detour can still bring a legged robot to the next stair or ledge a few centimetres or degrees off, and a fixed locomotion skill, factory or learned, can stall or fall.

- [Supporting source](https://arxiv.org/abs/2611.25788)

## Source-stated

The method records a completed crossing for each stair and ledge during one teaching session, then checks the approach after each local replan.

> for every stair and ledge, one crossing the skill has completed. Our method, passage-conditioned navigation, makes the route remember how each transition was entered

- [Supporting source](https://arxiv.org/abs/2611.25788)

## Source-stated

Under the reported localization-noise conditions, the method increased the share of disturbed ledge crossings within the specified limits from 24.1% to 88.6%.

> this raises the share of disturbed ledge crossings that stay within these limits from 24.1% to 88.6%.

- [Supporting source](https://arxiv.org/abs/2611.25788)

## Source-stated

In contact simulation, the authors report no falls in 180 climbs when the stored direction was used, compared with 9 falls when it was ignored.

> a robot pushed off course before stairs never falls in 180 climbs, against 9 falls when the stored direction is ignored

- [Supporting source](https://arxiv.org/abs/2611.25788)

## Source-stated

The authors report that their robot completed six competition runs over 271 m of multi-obstacle terrain and won third place.

> over 271 m of multi-obstacle terrain spanning 7.0 m of elevation, it completed all six runs, won third place at the 2026 Global Open-source AI Challenge

- [Supporting source](https://arxiv.org/abs/2611.25788)

## Source and contributor context

Sun, Wai Ka, Pong, Waiyiu, Wang, Bowen · Article author

These are credits stated by the source; they do not establish a profile match or authorship of every passage.


## Source media

No suitable media was found by the supported page extraction.


## Terms and references

Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.


## arXiv

arXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.

- [Explore this context](/entities/arxiv)
- [Reference](https://arxiv.org/)
- [Reference](https://info.arxiv.org/help/api/index.html)

## Editorial provenance

Version 1. Publication ID: 9a2c6638-6fe5-4cd8-846e-3433e6974a68.

Prepared by gpt-6-luna; grounding review by gpt-6-luna.

Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-20T07:18:51.376Z. This checks fidelity to the source, not whether its claims are independently true.

Editorial selection: accept; reasons: material\_development.

Observation: 9e7bf22174ddb4d500f11f6d9e59f38166e81177824d841b056c8d076df3a352. Source text hash: 935aafa9d763d9f4c8631c2aaf0ba370d032caa0e4432756d01fcf6a3e0498fb.


### uncertain — 2026-11-20T07:18:51.376Z — anthropic/claude-opus-5.5

Publication: 9a2c6638-6fe5-4cd8-846e-3433e6974a68

attribution — pass: Results and the motivating argument are consistently attributed to the authors ("The authors argue/report"). The inspection/patrol suitability is framed as a suggestion rather than an established fact, matching the source's own claim.

> This suits any robot with fixed skills that repeats a route over terrain transitions, as in inspection and patrol, where one wrong entry can end the task.

summary — uncertain: The reading omits the paper's central argument: existing methods need a robot model or extensive trial and error, which are unavailable on a new site, so the teaching session works without training or a robot model. Smallest repair: add one sentence stating this contrast. It also omits that the robot was the only fully autonomous finisher in its task. "Completed 180 climbs" slightly overstates "never falls in 180 climbs." Central terms (wheeled-legged, localization noise, local replan, gait schedule) need tooltips.

> Existing methods prevent this with a robot model or extensive trial and error, but on a new site with unknown skills neither is available, so they cannot be used.

date — pass: originalPublished 2026-11-19 matches the arXiv v1 submission date. The feed date of Nov 20 is later and correctly not used.

> \[Submitted on 19 Nov 2026\]

support — pass: The figures (24.1% to 88.6%, 180 climbs vs 9 falls, 271 m, six runs, third place), the gate thresholds and the 27-minute session match the abstract. The limitations and question appropriately note that the evidence is an abstract and that generalization is untested.

> before each ledge, an entry gate makes the robot turn or back up and retry if it is more than 10 cm or 10° off the taught crossing

media — pass: The selected media is the verified arXiv abstract page for this paper, which is the direct original source.

## Recorded references

Explicit citations in public readings to this reading. Shared topics are not citations. Other record types need an explicit reference adapter.

No public references recorded here yet.

[Reference data (JSON)](/readings/55b1ffeb77e27d3e027b9bb9c8512aca9a9fb7f5c7a34103a77cc4cd094d9355/references.json)

## Corrections and challenges

[Send a correction or an opposing source](https://escapemeridian.com/corrections/)

[JSON record](https://escapemeridian.com/readings/collision-free-is-not-enough-passage-conditioned-navigation-for-55b1ffeb.json)
