Aim Short to Reach Far: Your Frozen World Model Can Plan Better Than You Think

Aim Short to Reach Far: Your Frozen World Model Can Plan Better Than You Think

Xvyuan Liu, Jianjie Fang, Chen Gao, Yong Li · N/A · 2026

vision reinforcement-learning planning control human-robot-interaction benchmark

Framework

N/A

License

N/A

Stars

N/A

Summary

Planners built on visual world models commonly score each predicted outcome by its distance to the encoded goal image. We show that this target can limit control even with exact dynamics and globally optimal short-horizon search: reaching a goal may require actions that initially move away from it.

Abstract Summary

Planners built on visual world models commonly score each predicted outcome by its distance to the encoded goal image. We show that this target can limit control even with exact dynamics and globally optimal short-horizon search: reaching a goal may require actions that initially move away from it. With frozen LeWM models, intermediate targets substantially improve action synthesis and recorded-action ranking on Cube, PushT, Reacher, and TwoRoom. Learned targets and targets drawn from observed experience both produce these gains. We introduce Anchored Planning, which retrieves a recorded segment whose start and end resemble the current and goal observations, then aims at an observation shortly after its start. The frozen model scores actions toward this target from the current state. Without additional training, planning toward observed targets outperforms the released LeWM planner on every task in our long-range evaluation. Additional final-goal search falls short of the same gains. Lower successor-prediction error need not translate into better control. Success also depends on how far ahead the target is placed and on shrinking the retrieval span as execution advances. Changing only the target lets the same frozen model and planner reach goals that final-goal scoring misses.

Key Points

  • Planners built on visual world models commonly score each predicted outcome by its distance to the encoded goal image.
  • We show that this target can limit control even with exact dynamics and globally optimal short-horizon search: reaching a goal may require actions that initially move away from it.
  • With frozen LeWM models, intermediate targets substantially improve action synthesis and recorded-action ranking on Cube, PushT, Reacher, and TwoRoom.
  • Learned targets and targets drawn from observed experience both produce these gains
  • We introduce Anchored Planning, which retrieves a recorded segment whose start and end resemble the current and goal observations, then aims at an observation shortly after its start.

Why the Planning Target Matters

A frozen encoder $E$ maps current and goal images to $z_{t}=E(o_{t})$ and $z_{g}=E(o_{g})$. An action block $u$ contains $L=5$ successive primitive actions. The frozen predictor $F(z_{t},u)$ estimates the encoded state after that block and includes the fixed action normalization. The planner seeks the block whose predicted endpoint is closest to a target $q$: For a finite candidate set, we score every block.

For continuous actions, we approximate the optimization with the cross-entropy method (CEM) ( de Boer et al., 2005 ) .

Why the Planning Target Matters — Aim Short to Reach Far: Your Frozen World Model Can Plan Better Than You Think
Figure 1: Changing the target improves control with the same frozen predictor. (a) Both target rules score the same endpoints $p_{i}=F(z_{t},u_{i})$. Final-goal scoring selects $u_{1}$, whereas observed-target scoring selects $u_{2}$. Solid curves show predictions, and dashed segments show scoring distances. The target inset shows a recorded successor. Positions and observations are schematic. (b) On PushT, intermediate targets sustain control as the recorded goal offset grows. Curves use paired standard-start queries and the same frozen LeWM weights.

Anchored Planning

Figure 2 shows how Anchored Planning separates choosing a destination from choosing an action. We take an observed target, a waypoint on a recorded route toward the goal, from experience and use the frozen model to predict how candidate actions approach it from the current state.

Both memory settings below are built from the offline trajectories that accompany the world model, and no evaluation episode is ever stored in them.

Experiments

We organize the experiments around three questions about the effect of target choice, the role of search, and the design of useful targets: RQ1: How much control does changing the target reveal? RQ2: Can more search or longer lookahead recover the same gains? RQ3: What makes a target useful, and how does prediction help reach it?

We evaluate Cube, PushT, Reacher, and TwoRoom with task-specific pretrained LeWM models ( Maes et al., 2026 ) . Encoders, predictors, and the original action normalizers remain frozen.

Sources and Demonstrations

Figures are reproduced from Liu et al. See the full paper for experimental details and the project page, when the paper links one, for demonstrations and videos.

Share

Related Papers

Catch, Throw, Repeat: Planning for Human-Robot Partner Juggling
arXiv preprint

Catch, Throw, Repeat: Planning for Human-Robot Partner Juggling

Jonathan Rainer Lippert, Kai Ploeger, Abir Chowdhury et al.

Dynamic object exchange between humans and robots remains a challenging problem due to uncertainty in perception, timing, and contact-rich interaction.

vision reinforcement-learning planning control human-robot-interaction
PDF Intermediate
Has GPT-6-Astra Achieved Embodied Intelligence?
arXiv preprint

Has GPT-6-Astra Achieved Embodied Intelligence?

SenseNova, Joint Research Team, : et al.

GPT-6-Astra demonstrates a level of competence in robotic manipulation and navigation that challenges the conventional paradigm of building specialized models for individual tasks and embodiments.

manipulation locomotion vision reinforcement-learning planning control learning-from-demonstration benchmark
Code PDF Advanced
GitHub Stars:—