Riding the Shifting Potential: When Reactive Control Suffices for Multi-Goal Behavior

arXiv 2026 40.6 method

TLDR

Reactive control with graph-based world model and nullspace projections achieves multi-goal behavior, demonstrated in navigation and planar pushing with real robot transfer.

Reasoning

Strengths include a novel method for resolving objective conflicts via nullspace projections, strong empirical results (100% success) and real-world transfer. Weaknesses are limited domain scope and lack of comparison to other world model approaches.

Read-first score

Read-first score 40.6, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 12.

Recency 6%

100

Uses a gentle age decay so recent papers surface without erasing older foundations. 2026

Citation impact 18%

80.4

Uses OpenAlex-shaped citation metadata as a bibliometric attention signal, separate from paper quality. citation_normalized_percentile=0.80431093

Methodology quality 18%

Screens visible abstract and analysis fields for experiment, dataset, baseline, metric, and limitation evidence. markers=baseline

Reproducibility 18%

Screens links and visible text for paper, code, dataset, artifact, and repository signals. pdf=True; code=False; dataset=False; markers=code

Topical relevance 29%

17.1

Uses existing LLM keyword relevance scores normalized to 0-100. world model,world simulator,generative world model,interactive world model,video world model,world dynamics prediction,model-based reinforcement learning world model

Citation velocity 12%

Citation velocity estimates citations per publication-year to reduce old-paper bias. velocity=0.00

Field roles

FoundationFrontierBridge

Rank sensitivity

Stability: volatile; rank range: 266.

Keyword Scores

world model

interactive world model

world simulator

world dynamics prediction

generative world model

video world model

model-based reinforcement learning world model

Deep Analysis

Innovations

Arguing that reactive control is sufficient for multi-objective tasks when using dynamic priority encoding via nullspace projections on a graph-based world model
Resolving conflicts by projecting lower-priority gradients into nullspace of higher-priority ones, with priorities determined continuously from current state
Demonstrating 100% success on planar pushing of non-convex objects without demonstrations or retraining, outperforming steepest-descent (0%) and diffusion policy (~55%)
Direct transfer to real robot with additional perceptual and kinematic constraints using the same mechanism

Methodology

The method extends a graph-based world model with nullspace projections to resolve conflicts between objectives. Priorities are determined continuously from the current state, and lower-priority gradients are projected into the nullspace of higher-priority ones. The approach is evaluated on navigation around non-convex obstacles and planar pushing of non-convex objects, comparing against steepest-descent baseline and diffusion policy.

Key Results

On planar pushing of non-convex objects, the method achieves 100% success across one-hundred configurations versus 0% for steepest-descent baseline and approximately 55% for diffusion policy, without demonstrations or retraining. The same formulation transfers directly to a real robot with additional constraints.

Resources

Paper