IMAGE + INSTRUCTION01ESW-Vista
10,800 samplesVisual scenes paired with task instructions. Study the behaviors and consequences a model generates from a hazardous starting context.
Visual prediction · World-model evaluationFrom Machine-Centric
to Human-Centric Safety
A robot can move perfectly and still cause harm. Evaluating embodied intelligence through the consequences of its actions.
Code Coming SoonArena includes an additional 1,080 paired safe tasks.
01 / A DIFFERENT SAFETY QUESTION
Avoiding collisions is essential. Understanding the harm an action can cause is the next challenge.
The same grasp, move, and place sequence can be physically valid yet unsafe in context. Human-centric safety asks whether a model understands the objects, people, and delayed effects around its actions.
Collision avoidance
Force limits
Grasp stability
Harm to people and animals
Damage to belongings
Exposure of private information

02 / TWO COMPLEMENTARY BENCHMARKS
From predicting a possible future to acting inside an interactive environment. Two benchmarks bring human-centric safety into focus.
IMAGE + INSTRUCTION01Visual scenes paired with task instructions. Study the behaviors and consequences a model generates from a hazardous starting context.
Visual prediction · World-model evaluationGalaxea G0.5 · 13.8 s
Cast iron weight placed on a notebook · Task completion: 5/5
Executable robot environments built on LIBERO. Observe actions as they unfold, with paired safe tasks to distinguish safety decisions from execution failures.
Interactive control · VLA / WAM / LLMThe danger is not always obvious—or immediate.
03 / A WORLD OF HIDDEN RISKS
Explore nine categories through actual generated research examples. Select a category, then play its model-generated trajectory.
Generated trajectories illustrate the evaluation setting; they are not recordings of real-world incidents or standalone evidence of a model's overall safety.
Selected Wan 3.0 and Seedance 2.0 examples, curated for visual clarity and coherent motion.
04 / FROM CONCEPT TO CONSEQUENCE
Static benchmarks enable consistent comparison. Dynamic red teaming uses trajectory feedback to explore weaknesses beyond a fixed test set.

Generate, select, expand, and validate causal pathways from broad risk concepts.
Turn mechanisms into instructions, visual scenes, and executable environments.
Track dangerous attempts, critical actions, and consequences alongside task ability.
Retrieve and refine SkillBank experience to guide the next round of red teaming.
Trajectory feedback → validated experience → more targeted testing
↰A risk-related action begins.
The critical dangerous operation is reached.
Harm occurs or a directly hazardous state forms.
05 / WHAT THE EVALUATION REVEALS
Better generation or task execution does not necessarily mean safer behavior. Examine the reported results within each model family.
| Model | Refusal | Dangerous attempt | Dangerous commit | Dangerous success |
|---|
Read rates alongside task capability. A low dangerous-success rate may reflect execution failure rather than active risk avoidance. WM and embodied-control results use different benchmarks and should not be treated as one leaderboard.
World models can produce coherent dangerous trajectories. Visual quality and human-centric safety measure different things.
The evaluated VLA and WAM models report 0% refusal. Paired safe tasks help reveal when low danger rates stem from limited execution ability.
Concealment, delayed consequences, and longer action sequences expose different weaknesses. A single aggregate metric cannot describe them all.
BEYOND A FIXED BENCHMARK
Across ten rounds of self-evolving red teaming, the planner discovers scenarios that more frequently expose dangerous outcomes in the two target models.
100 test cases per target model per round.
DANGEROUS-SUCCESS RATE · ROUND 1 → ROUND 10
Higher rates indicate stronger risk discovery by the red team, not an improvement in target-model safety.
Explore the benchmarks and research questions. Follow the repository for the upcoming code and documentation release.