Unsupervised Learning of Goal Spaces for Intrinsically Motivated Goal Exploration
This work addresses a limitation in autonomous skill acquisition for robots, though it is incremental as it builds on existing exploration algorithms by replacing engineered spaces with learned ones.
The paper tackles the problem of enabling intrinsically motivated goal exploration algorithms to autonomously sample goals without relying on engineered feature spaces, by using deep representation learning to learn a goal space from raw sensor observations, and shows that this approach matches the performance of engineered representations in simulated robot arm experiments.
Intrinsically motivated goal exploration algorithms enable machines to discover repertoires of policies that produce a diversity of effects in complex environments. These exploration algorithms have been shown to allow real world robots to acquire skills such as tool use in high-dimensional continuous state and action spaces. However, they have so far assumed that self-generated goals are sampled in a specifically engineered feature space, limiting their autonomy. In this work, we propose to use deep representation learning algorithms to learn an adequate goal space. This is a developmental 2-stage approach: first, in a perceptual learning stage, deep learning algorithms use passive raw sensor observations of world changes to learn a corresponding latent space; then goal exploration happens in a second stage by sampling goals in this latent space. We present experiments where a simulated robot arm interacts with an object, and we show that exploration algorithms using such learned representations can match the performance obtained using engineered representations.