Back to the Familiar Future: Failure Recovery for VLA Policies via Pre-Imagined Milestone Selection
Back to the Familiar Future: Failure Recovery for VLA Policies via Pre-Imagined Milestone Selection
This paper introduces Back to the Familiar Future (B2FF), a recovery framework for vision-language-action (VLA) policies that uses pre-imagined milestone states conditioned on clean initial observations. At runtime, a recoverability-aware selector chooses a familiar future state from a generated bank and enforces it as a fixed visual goal to guide the policy back from off-trajectory observations. Evaluated on failure-injected LIBERO with recovery timing aligned to injected failures, B2FF increases the average success rate of a baseline VLA from 56.3% to 74.0% without fine-tuning the low-level action generator.
The study presents a framework called Back to the Familiar Future (B2FF) that improves failure recovery for vision-language-action (VLA) policies by selecting pre-imagined milestone states; on the LIBERO benchmark with injected failures, it raises average success rates from 56.3% to 74.0% without fine-tuning the low-level action generator.
Source: arXiv Robotics — research abstracts · Read original article ↗
Loading article text…
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
Reported numbers
Reported success rate
56.3%
Reported trials
View original evidence
ations back to a familiar future. On failure-injected LIBERO, under controlled recovery timing aligned with the injected failure, B2FF increases the average success rate of a baseline VLA from 56.3% to 74.0%, demonstrating that pre-imagined milestones can guide recovery without fine-tuning the low-level action generator.
Open source S4Reported success rate
74%
Reported trials
View original evidence
ations back to a familiar future. On failure-injected LIBERO, under controlled recovery timing aligned with the injected failure, B2FF increases the average success rate of a baseline VLA from 56.3% to 74.0%, demonstrating that pre-imagined milestones can guide recovery without fine-tuning the low-level action generator.
Open source S4
- paper: not_reported
What remains unknown
Not established in the collected evidence: Environment, Control, Data origin.
Reported performance applies to the described task. It does not establish general autonomy or deployment readiness.
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
ations back to a familiar future. On failure-injected LIBERO, under controlled recovery timing aligned with the injected failure, B2FF increases the average success rate of a baseline VLA from 56.3% to 74.0%, demonstrating that pre-imagined milestones can guide recovery without fine-tuning the low-level action generator.
Open source S4
arXiv:2606.09258v2 Announce Type: replace Abstract: Vision-language-action (VLA) policies can deviate from nominal trajectories during manipulation, even when tasks remain physically feasible. Recovering from these deviations is challenging, as they push the policy into unfamiliar state spaces where direct re-planning frequently destabilizes action sequences. We propose Back to the Familiar Future (B2FF), a recovery
Open source S2
Back to the Familiar Future: Failure Recovery for VLA Policies via Pre-Imagined Milestone Selection
Open source S1
framework for foresight-driven VLAs that leverages future visual conditioning as a recovery interface. Before execution, the VLA generates a milestone bank of familiar future states conditioned on the clean initial observation. At recovery time, a recoverability-aware selector selects a recovery milestone from this bank and enforces it as a fixed visual goal. This enables the VLA to robustly map off-trajectory observ
Open source S3
Source：arXiv Robotics — research abstracts · arxiv.org