GroundedPlanBench: Spatially grounded long-horizon task planning for robot manipulation

1. 1Identify the limitations of current vision-language models in planning robot actions
2. 2Develop a spatially grounded approach to integrate planning and execution
3. 3Evaluate the performance of GroundedPlanBench in long-horizon task planning for robot manipulation

Robotics and AI engineers on a team can benefit from this research as it improves the planning and execution of robot actions, while product managers can consider its applications in real-world scenarios

💡 Integrating planning and execution can improve the performance of vision-language models in robot manipulation tasks

🤖 GroundedPlanBench: a new approach for spatially grounded long-horizon task planning in robot manipulation