Generating Adaptive Robotic Behaviours via Enhanced Diffusion Policy
Tianmei Jin, Jiayi Zhang · 2024
While manual robot programming has been effective for many applications, fixed coding poses several challenges. Robot programming requested sophisticated and dynamic behaviours while increasing the complexity of the robot's tasks. Generative artificial intelligence models have revolutionised robot behaviour generation in dynamic environments to complete different tasks. This paper explores different approaches to robot behaviour generation evaluating their effectiveness, challenges, and potential implications for real-world robotic scenarios. An enhanced diffusion policy is proposed to mitigate anomalous behaviours in the original model. The results demonstrate the importance of training dataset quality and model adaptation to specific working environments in achieving successful robotic behaviours. The Resilient Diffusion solved unusual behaviour problems, improved the resilience capability of diffusion policy, and achieved a higher success rate.