Bounding Boxes as Goals: Language-Conditioned Grasping via Neuro-Symbolic Planning
Researchers have developed a new framework called GRASP (Grounded Reasoning and Symbolic Planning) that enables robots to understand and execute natural language commands for manipulation tasks. This neuro-symbolic approach uses a pre-trained vision-language model to translate abstract queries into physical goal states, identified by bounding boxes. GRASP achieves a 73.3% success rate in real-world trials without requiring task-specific training, demonstrating its potential for open-vocabulary tabletop manipulation. AI
IMPACT This framework could significantly advance robot autonomy by enabling more intuitive, language-based control for manipulation tasks.