Simulation plays a key role in scaling robot learning and validating policies, but constructing simulations remains labor-intensive. In this paper, we introduce \ourmethod{}, a generative simulation framework that automates this process using inverse design. Given an agent's behavior (such as a motion trajectory or objective function) and its textual description, we infer the underlying scenarios and environments that could have caused the behavior.
Our approach leverages large language models to construct and expand a graph that captures cause-and-effect relationships and relevant entities with properties in the environment, which is then processed to configure a robot simulation environment. Our approach supports (i) augmenting simulations based on ego-agent behaviors, (ii) controllable, counterfactual scenario generation, (iii) reasoning about agent cognition and mental states, and (iv) reasoning with distinct sensing modalities, such as braking due to faulty GPS signals.
We demonstrate our method in autonomous driving and robot manipulation tasks, generating more diverse, complex simulated environments compared to existing simulations with high success rates, and enabling controllable generation for corner cases. This approach enhances the validation of robot policies and supports data or simulation augmentation, advancing scalable robot learning for improved generalization and robustness.
Crossing with a malfunctioned traffic light
Advancing after distracted driver moves
Waiting for intersection to clear
Yielding for ambulance at intersection
Proceeding after pedestrian crosses street
Block outside odor
Protect plant from direct sunlight
Trap candle's scent
Keep out unwanted gaze
Block out construction noise
Slowing for lane closure congestion
Yielding to overtaking ambulance
Maintaining wary distance behind stopping truck
Slowing to avoid SUV cut-in
Complying with police speed enforcement
Clean dirty dishes
Thaw frozen vegetables
Water the plant
Washing mixed fruits in a bowl
Collect tap water
Avoiding open door hazard
Avoiding collision with elderly pedestrian
Stopping abruptly due to highway accident
Yielding safely to passing ambulance
Braking for roadside assistance hazard
Improve air circulation and ventilation
Bid farewell and let guest out
Check delivered package
Braking for occluded SUV at corner
Stopping at a walking street
Halting at unexpected police checkpoint
Yielding to ambulance
Picking up a passenger
1. Open the storage furniture
Store detergent out of children's reach
2. Put the detergent away
3. Close the storage furniture door
1. Open the storage furniture
Put away toys as guest arrive
2. Put away toys
3. Close the storage furniture door
Yielding to ambulance
Braking to avoid running a red light
Stopping for a car running a red light
Brake for a fallen box
Halt for a high-speed police chase
Showcase an artwork
Brighten the desk to read a book
Setup workstation
Abrupt lane change to avoid debris
Overtaking slow vehicle
Clearing path for ambulance with lane change
Merging to open lane
Swerving to avoid wrong-way driver
Remove glare on the screen
Sharing a presentation during a meeting
Grabbing a bucket to fill with sand
Fetching water for floor mopping