Awesome World Model Hub Papers · Datasets · Projects
← Back to papers

SWAP: Symmetric Equivariant World-Model for Agile Robot Parkour

arXiv 2026 53.5 method

TLDR

SWAP embeds symmetry into a world model for quadruped parkour, achieving record-breaking jumps and zero-shot generalization.

Reasoning

The paper presents a novel symmetric equivariant world model that reduces learning burden and achieves impressive real-world parkour records. Strengths include real-world validation and strong geometric generalization, but the approach is limited to symmetric quadruped locomotion and lacks explicit baseline comparisons in the abstract.

Read-first score

Read-first score 53.5, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 35.

Recency 6%
100

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

Citation impact 18%
88.4

Uses OpenAlex-shaped citation metadata as a bibliometric attention signal, separate from paper quality. citation_normalized_percentile=0.88370536

Methodology quality 18%
60

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

Topical relevance 29%
50

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

Reproducibility 18%
38

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

Citation velocity 12%
0

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

Field roles

FoundationFrontierBridge

Rank sensitivity

Stability: volatile; rank range: 343.

Keyword Scores

world model
10
model-based reinforcement learning world model
9
world dynamics prediction
7
world simulator
5
generative world model
3
interactive world model
1
video world model
0

Deep Analysis

Innovations

  • End-to-end equivariant symmetric world model that embeds left-right symmetry directly into both the world model and actor-critic networks
  • Reduction of learning burden by avoiding redundant encoding of symmetric interactions in latent world models
  • Demonstration that symmetry equivariance serves as an effective structural prior for pushing the physical boundaries of learned legged locomotion

Methodology

SWAP is an end-to-end equivariant symmetric world model that incorporates symmetry directly into the world model and actor-critic networks. It uses a latent world model architecture with equivariant layers to enforce left-right symmetry, reducing redundant learning. The model is trained on real-world robot data and evaluated on parkour tasks including gap crossing and platform climbing.

Key Results

The robot successfully leaps across a 2.13 m gap and climbs a 1.63 m platform, breaking previous records for quadruped parkour. The framework also demonstrates robust geometric generalization to unseen mirrored terrains and zero-shot transferability across diverse outdoor environments.

Tags

equivariant world modelquadruped parkoursymmetryreinforcement learninggeometric generalizationRO