Socially aware AI agents should be able, among other things, to collaborate fluently with a human in tasks that require interdependent action in order to be solved. Towards enhancing mutual performance, collaborative AI agents should be equipped with adaptation and learning capabilities. However, co-learning requires long training intervals so that both partners learn and adapt to each other. To alleviate this, transfer learning methods could be explored to shorten training and improve performance. In the current thesis, we studied the experience and performance of human-agent teams in a task where a human and a Deep Reinforcement Learning (DRL) Soft-Actor-Critic (SAC) agent needs to learn in real-time how to collaborate in order to achieve a common goal. To test the benefits of transfer learning, a Learning from Demonstration method was used that utilized demonstration data from a human-agent expert team to facilitate the co-learning procedure. The proposed methods were evaluated through a study with 8 different human-agent teams, half of which played the game without transfer learning, while the rest with transfer learning. The results indicate that applying transfer learning in scenarios where the agent needs to collaborate with different humans has the potential to shorten training duration and improve the overall experience.