❓ Did you know that research on #autonomousvehicles started three decades ago. Crazy, eh?
🏆 An intelligent agent is an entity that can autonomously act in an environment to achieve its goals. Reinforcement Learning (RL) is one way in which these agents can learn the optimal action to be taken in a specific situation. However, RL needs a means to provide rewards for the agent’s actions, and it is not always easy to define reward functions, especially in real-world applications. What can we do in these cases?
🦜Did you know about Imitation Learning (IL)? These are useful algorithms for scenarios where it is difficult to define a reward function.
🔸In IL, instead of trying to learn from the rewards or manually specifying a reward function, an expert (typically a human) provides the agent with a set of demonstrations.
🔸The agent then tries to learn the optimal policy -- a mapping between observations (states) and actions -- by following or imitating the expert’s actions or decisions in the demonstrations.
🔸IL enables us to teach robots to perform actions or make decisions without explicitly writing code for it, but by providing only a small number of #kinesthetic or visual demonstrations.
🔸#ALVINN was trained to predict the “turn curvature” to be taken by the vehicle in order to follow the road. The training was done using road images captured from a driving simulation.
🔸To learn more about imitation learning for #robotics, check out this course material:
🔸Some interesting videos:
History of Autonomous Cars
Imitation learning techniques and recent applications