{"object":"list","default_model":"kine-a1","data":[{"id":"kine-a1","name":"Kine-A1","object":"model","owned_by":"kinejing","status":"experimental","endpoint":"/v1/actions","trainable_parameters":833694,"checkpoint_sha256":"504d2d85529701a06a73ddf9b2d7ec291d3fb125a479fb83899ec0e7118b724a","inference_asset_sha256":"7f3a04808c62ea85991f32be76352f18905199a58802e9c55fe73073e981e193","input":"two RGB256 PNG images, native state8, exact supported task","supported_task":"Pick up the black bowl between the rectangular board and the small ramekin, and place it on the plate.","image_orientation":"robosuite_raw_unflipped","state_schema":"robosuite-eef-pos-axis-angle-gripper-qpos-8d-v1","action_schema":"robosuite-osc-pose-gripper-7d-v1","action_shape":[50,7],"simulation_only":true,"commercial_ready":false,"task_quality_verified":false,"limitations":["仅支持指定仿真场景和固定任务；不理解任意任务文本。","最近一次中止评估的首个完整场景失败；没有有效的完整成功率结论。","返回动作不代表任务完成，不可直接用于真实机器人。","不生成视频、3D世界或未来画面；不承诺通用世界模型能力。"]},{"id":"kinejing-state-pusht-v1","owned_by":"KineJing","status":"research_preview","commercial_ready":false,"output_kind":"predicted_simulator_states","state_schema":"pusht-512xy-angle-rad-v1","action_schema":"pusht-absolute-xy-10hz-v1","state_fields":["agent_x_px","agent_y_px","block_x_px","block_y_px","block_angle_rad"],"action_fields":["absolute_agent_target_x_px","absolute_agent_target_y_px"],"context_states":2,"state_dimensions":5,"action_dimensions":2,"min_horizon":1,"max_horizon":16,"validated_horizons":[1,3,16],"time_step_seconds":0.1,"coordinate_frame":"512x512 native PushT simulation; angles modulo 2pi","action_target_range_px":[0,512],"training_action_target_range_px":[32,480],"inference_asset_sha256":"92a0b80a535baa1e43015692a9c39a3b34d47aef3b45559f93faf2509ed47549","protocol_sha256":"c82c980602192c40d02825fc257ccdfc7262a6566c30b0ccd26beb363056267b","evaluated_runtime_sha256":"32a47109d662d2887108f4d87e22c90a4de9d119f4842c08bd3529001161b380","trainable_parameters":417042,"parameters_per_member":139014,"ensemble_members":3,"external_pretrained_parameters":0,"external_encoder_required":false,"training_episodes":256,"validation_episodes":32,"test_episodes":32,"test_scope":"new heldout episodes from the same simulation and exploration distribution","test_3step":{"block_xy_rmse_px":9.542668082161342,"block_angle_rmse_rad":0.10593133506446288,"ridge_block_xy_rmse_px":16.100442816882047,"persistence_block_xy_rmse_px":24.63955673098502,"block_encoded_error":0.005793879203338292,"ridge_block_encoded_error":0.020097255750442855,"candidate_minus_ridge_ci95":[-0.017410279193518362,-0.011470266102651295]},"scope":"First-party networks trained from random initialization to predict 2D PushT object-state consequences under supplied actions.","limitations":["Requires two complete agent/object poses 0.1 seconds apart; no RGB input or video output.","Small domain model; no general foundation-model, real-robot, planning-success or safety claim.","Test uses the same simulator and exploration distribution; no OOD validation.","Recorded evaluation actions come from a state-aware exploration policy; this is forced-action offline prediction.","Pose history omits hidden velocities/contact forces; rollout errors accumulate.","The ridge baseline predicts directly actuated pusher XY more accurately.","Action targets outside the training range [32,480] are unvalidated."],"provenance":{"training_data":"company self-collected simulation trajectories; no downloaded benchmark data","simulation_repository":"https://github.com/real-stanford/diffusion_policy","simulation_commit":"5ba07ac6661db573af695b419a7947ecb704690f","simulation_license":"MIT","research_evidence":"research/state-pusht-v1"}},{"id":"kinejing-dynamics-v0.1","status":"research_preview","commercial_ready":false,"output_kind":"predicted_latent_features","output_shape":[9,3,768],"views":["head","left","right"],"feature_space":"dinov2-vits14-cls-meanpatch-v1","encoder_weights_sha256":"b938bf1bc15cd2ec0feacfe3a1bb553fe8ea9ca46a7e1d8d00217f29aef60cd9","checkpoint_sha256":"f9cf4aeac48ae603c8eedf9f8eb81aa33f939b49406f078f12e2ca9a9c249166","inference_asset_sha256":"9bf206671001259377dcd0635af9a9886172b66e9ea453c41c921ccd8f1e4a7d","trainable_parameters":1341824,"input_normalization":"raw_features_and_dataset_native_actions_normalized_by_training_statistics","action_schema":"triworldbench-joint-action-vector-v1","action_rows":9,"action_dimensions":14,"time_grid":[0.0,0.125,0.25,0.375,0.5,0.625,0.75,0.875,1.0],"time_grid_contract":"Nine increasing normalized source-frame times, first 0 and last 1; defaults to equal spacing. Preserve original rounded frame indices divided by source frame count minus one when available.","scope":"Trained three-view action-conditioned DINOv2 feature predictor; no RGB video, robot control, or general world-model claim.","feature_extraction":{"architecture":"dinov2_vits14","repository":"https://github.com/facebookresearch/dinov2","revision":"7764ea0f912e53c92e82eb78a2a1631e92725fc8","preprocessing":"OpenCV linear resize RGB 224x224; ImageNet mean/std","representation":"concatenate CLS token (384) and mean patch token (384)","encoder_included":false}}]}