diff --git a/projects/ism2026_ros2smolvla.md b/projects/ism2026_ros2smolvla.md
index 4d5aeb1..10a91e8 100644
--- a/projects/ism2026_ros2smolvla.md
+++ b/projects/ism2026_ros2smolvla.md
@@ -147,11 +147,11 @@ We present ROS2SmolVLA, a ROS 2-based interface that enables the deployment of H
 Training and evaluation data sets, and model parameters are available in our Hugging Face collection [ROS2SmolVLA](https://huggingface.co/collections/una-auxme/ros2smolvla).
 
 **Software**
-- **ROS 2 Interface ([LeRobot-ROS2 interface](https://github.com/una-auxme/ros2smolvla_interface_lerobot)):** Acts as a lightweight broker between the LeRobot ecosystem and ROS 2, enabling seamless communication between the VLA policy and robot controllers without managing low-level hardware cycles.
-- **Multi-Modal Data Acquisition ([LeRobot-camera interface](https://github.com/una-auxme/ros2smolvla_interface_camera)):** Standardizes ingestion of asynchronous camera feeds from top, side, and end-effector-mounted cameras, converting them into uncompressed images compatible with SmolVLA’s input requirements.
-- **Containerized Inference ([UR10e VLA Docker](https://github.com/una-auxme/ros2smolvla_docker)):** Isolates dependencies and ensures reproducibility across development, simulation, and production environments, enabling efficient execution on edge devices with consumer-grade hardware.
-- **Simulation Environment ([UR10e simulation in Gazebo](https://github.com/una-auxme/ros2smolvla_ur10e_sim)):** A Gazebo-based digital twin that replicates the physical robot’s kinematics, camera setup, and workspace layout, enabling pre-validation and policy testing in a simulated environment before real-world deployment.
-- **Real-World Data Collection ([UR10e interface](https://github.com/una-auxme/ros2smolvla_ur10e_real)):** A dedicated ROS 2 package for capturing high-fidelity, real-world training data from the physical UR10e setup, including synchronized joint states, camera streams, and action trajectories—essential for robust, sim-to-real transfer and model fine-tuning.
+- **ROS 2 Interface ([ros2smolvla_interface_lerobot](https://github.com/una-auxme/ros2smolvla_interface_lerobot)):** Acts as a lightweight broker between the LeRobot ecosystem and ROS 2, enabling seamless communication between the VLA policy and robot controllers without managing low-level hardware cycles.
+- **Multi-Modal Data Acquisition ([ros2smolvla_interface_camera](https://github.com/una-auxme/ros2smolvla_interface_camera)):** Standardizes ingestion of asynchronous camera feeds from top, side, and end-effector-mounted cameras, converting them into uncompressed images compatible with SmolVLA’s input requirements.
+- **Containerized Inference ([ros2smolvla_docker](https://github.com/una-auxme/ros2smolvla_docker)):** Isolates dependencies and ensures reproducibility across development, simulation, and production environments, enabling efficient execution on edge devices with consumer-grade hardware.
+- **Simulation Environment ([ros2smolvla_ur10e_sim](https://github.com/una-auxme/ros2smolvla_ur10e_sim)):** A Gazebo-based digital twin that replicates the physical robot’s kinematics, camera setup, and workspace layout, enabling pre-validation and policy testing in a simulated environment before real-world deployment.
+- **Real-World Data Collection ([ros2smolvla_ur10e_real](https://github.com/una-auxme/ros2smolvla_ur10e_real)):** A dedicated ROS 2 package for capturing high-fidelity, real-world training data from the physical UR10e setup, including synchronized joint states, camera streams, and action trajectories—essential for robust, sim-to-real transfer and model fine-tuning.
 
 ## Results