HuggingFaceVLA

/

smolvla_libero

like 15

Follow

Hugging Face Vision Language Action Models Research 104

I evaluated the model on the lerobot/libero dataset using episode 0 for a 100-step inference. For each step, I converted the predicted delta end-effector (EE) poses into absolute target EE poses. I then computed the joint angles using the kdl.ChainIkSolverPos_LMA inverse kinematics solver and commanded them to the Gazebo simulation. However, the simulated robot did not replicate the trajectories shown in the dataset video.

urdf file i use is below

false true false false false false false false false false false false false false false 1.7453 256 256 0.01 100 1 30 true camera_top/image /media/vijayrana/c792b065-45ed-465c-ba30-ea7094e05f12/vijay/Work/gazebo_openarm_sim/install/franka_gazebo/share/franka_gazebo/config/franka_controllers.yaml gz_ros2_control/GazeboSimSystem 0.0 -0.785 0.0 -2.356 0.0 1.571 0.785 0.0