G1 depth dodgeball on frozen SONIC (ONNX deploy)

Frozen SONIC whole-body controller + seven rank-16 LoRAs on the decoder, conditioned on a 128-d percept from a head depth camera (Theia-Tiny tokens + temporal adapter). PPO, iteration 4000. Consumed by lerobot's DepthDodgeController for the Unitree G1.

file	inputs	outputs
`theia_image.onnx`	`image` uint8 `[B,224,224]` (depth mapped 255 near .. 1 far, 0 invalid)	`tokens` `[B,197,192]` fp32
`perception.onnx`	`depth_tokens` fp32 `[1,16,197,192]` (16-frame ring at 25 Hz, fp16-rounded)	`percept` `[1,128]`
`actor.onnx`	`tokenizer[1,640]`, `policy[1,930]`, `conditioning[1,768]` = concat(tokenizer, percept)	`actions[1,29]` SONIC-normalized PD residuals, MuJoCo joint order

Camera contract: 64x96 optical-Z depth in metres, 45 deg vertical FOV, 0 deg pitch, valid 0.2-6 m, 25 Hz. See manifest.json for the full training config.

Downloads last month: -; Downloads are not tracked for this model. How to track

Inference Providers NEW

This model isn't deployed by any Inference Provider. 🙋 Ask for provider support