Lottie Env

Agent receives 3 reference keyframes → generates Lottie JSON → reward based on MSE similarity.

Task Flow

POST /reset
→
Observation: 3 reference frames
→
POST /step with Lottie JSON
→
Reward [0,1] + submitted frames

Easy Task: Bouncing Ball Easy

Reference Keyframes

Start frame
frame_start
Middle frame
frame_middle
End frame
frame_end

Agent Output

Medium Task: Work in Progress Medium

Reference Keyframes

Start frame
frame_start
Middle frame
frame_middle
End frame
frame_end

Agent Output

Hard Task: Saving the Money Hard

Reference Keyframes

Start frame
frame_start
Middle frame
frame_middle
End frame
frame_end

Agent Output

Schemas

FieldType
LottieAction.lottie_jsonstring (complete Lottie JSON)
LottieObservation.{start,middle,end}_framePNG image (reference)
LottieObservation.submitted_*_framePNG image (agent's)
LottieObservation.rewardfloat in [0, 1] or -1.0
LottieObservation.donebool (always false)
API Docs Schemas Health