More broadly, our results suggest that simulation has been held back not by the idea of simulation, but by unrealistic simulations, and that coding agents can now build realistic ones in which far more capable frontier models learn in ways that the real world does not allow.
Methodology
This section introduces the problem setup and the GDP-grounded selection of simulations. It also describes the multi-agent strategy for creating simulations and how tasks of controlled difficulty are created for each simulation.
Scaling Tasks
To our knowledge, there is no established definition of a good task. We therefore define six criteria that a task must meet to be useful for evaluating and training robot policies:
- Meaning. The task is a realistic and meaningful job, described naturally as one would describe it to a person, not as a checklist of verifier conditions.
- Setting. The scene, objects, and arrangement suit the specific task. For instance, a free kick that curls the ball requires a defensive wall, and a dribbling task requires cones.
- Verifier correctness. The verifier is deterministic and computed from the physical state of the simulation, without a VLM judge. It has neither false positives, such as accepting a cup placed over a ball for a task of putting the ball in the cup, nor false negatives, such as requiring the ball to rest at the exact center of the cup when the task did not ask for it.
- Solvability. As for simulations, assisted demonstrations show that every interaction required by the task is physically achievable, without requiring an end-to-end solution.
- Reward quality. The reward gives continuous partial credit that reflects meaningful progress, and full credit to every valid solution, irrespective of the strategy used. For instance, a cleanup task credits each object placed in its correct location.
- Difficulty coverage. Tasks that are all too easy or all too hard provide little signal for evaluating or improving models. We therefore measure the difficulty of each task with a weaker and a stronger model, Claude Haiku 5.5 and Claude Opus 5.5, and target a spread of success rates for both. This gives a ladder of difficulty that remains informative for both weaker and stronger models.
Learning in Simulation with Frontier Models
Realistic simulations open up a new way to learn physical skills: practicing in simulation, without the cost and risk of a real robot. Prior work mostly uses simulation to train small policies with reinforcement learning, often over millions of trials per task. Frontier models can learn in a different way, since they can reason about a task, write code, and improve from feedback within a single context. We therefore study how frontier models can learn in simulation, and how what they learn can be passed to the models and policies that control a real robot.
Since a simulation can be reset instantly, a frontier model can attempt the same task repeatedly and improve from its own failures within a single context, without any weight updates. After each attempt, the simulation is reset to the same initial state, and the agent continues in the same conversation and workspace, retaining its observations, code, and notes from earlier attempts. We run three attempts of three hours each. To measure the value of the ground-truth state, we compare agents with restricted access against agents with privileged access, which can additionally read the simulator state during each attempt.
Conclusion
More broadly, our results suggest that simulation has been held back not by the idea of simulation, but by unrealistic simulations, and that coding agents can now build realistic ones in which far more capable frontier models learn in ways that the real world does not allow. We hope that Gym-Anything for Robotics, RoboWorld, and the released code provide a foundation for future work on realistic robot learning, including more capable embodied agents, new ways to learn from privileged state, and broader coverage of physical work.
Frequently Asked Questions
What criteria define a useful task? Tasks must meet six criteria: meaning, setting, verifier correctness, solvability, reward quality, and difficulty coverage.
How is task difficulty measured? The authors assess each task with Claude Haiku 5.5 and Claude Opus 5.5, targeting a spread of success rates for both models.
How do frontier models practice in simulation? They make repeated attempts from the same initial state, retaining observations, code, and notes from previous attempts in the same conversation and workspace.
How many practice attempts are run, and how long is each? The study runs three attempts, each lasting three hours.