Andrea Fumagalli

Ashit Kumar Subudhi, Bhargav Chirumamilla, Shubham Vaishnav et al.

Dynamic traffic variations in Open Radio Access Networks (O-RAN) lead to drift, which degrades the performance of Artificial Intelligence/Machine Learning (AI/ML) models. Traditional retraining approaches maintain forecasting accuracy but incur high computational cost and may lead to violations of Service Level Agreements (SLAs). This work proposes a Q-learning-based adaptive retraining approach that formulates the retraining decision as a Markov Decision Process (MDP), where a Reinforcement Learning (RL) agent learns a policy that balances forecasting accuracy and retraining cost. The proposed approach incorporates a multi-expert Long Short-Term Memory (LSTM) ensemble to mitigate catastrophic forgetting and improve robustness across diverse traffic conditions. Experimental results show that the proposed approach effectively reduces retraining overhead compared to greedy and random baselines, while maintaining system performance within predefined limits.

Yuan Tian, Lianjun Li, Andrea Fumagalli et al.

Robots assist in many areas that are considered unsafe for humans to operate. For instance, in handling pandemic diseases such as the recent Covid-19 outbreak and other outbreaks like Ebola, robots can assist in reaching areas dangerous for humans and do simple tasks such as pick up the correct medicine (among a set of bottles prescribed) and deliver to patients. In such cases, it might not be good to rely on the fully autonomous operation of robots. Since many mobile robots are fully functional with low-level tasks such as grabbing and moving, we consider the mixed-initiative control where the user can guide the robot remotely to finish such tasks. For this mixed-initiative control, the user controlling the robot needs to visualize a 3D scene as seen by the robot and guide it. Mixed reality can virtualize reality and immerse users in the 3D scene that is reconstructed from the real-world environment. This technique provides the user more freedom such as choosing viewpoints at view time. In recent years, benefiting from the high-quality data from Light Detection and Ranging (LIDAR) and RGBD cameras, mixed reality is widely used to build networked platforms to improve the performance of robot teleoperations and robot-human collaboration, and enhanced feedback for mixed-initiative control. In this paper, we proposed a novel haptic-enabled mixed reality system, that provides haptic interfaces to interact with the virtualized environments and give remote guidance for mobile robots towards high-level tasks. The experimental results show the effectiveness and flexibility of the proposed haptic enabled mixed reality system.