Time Reversal as Self-Supervision
Suraj Nair, Mohammad Babaeizadeh, Chelsea Finn, Sergey Levine, Vikash Kumar
- Collects trajectories exploring outward from a set of goal states and reverses them.
- Trains a supervised model, the time reversal model (TRM), to predict these reversed trajectories.
- Uses the trained TRM to predict the trajectory of states leading to the goal state for a new scene.
- Uses cross entropy method with a learnt forward model to follow the trajectory predicted by TRM in visual model predictive control fashion.
Generalizing to varying initial conditions, diverse objects, and changing objectives has been a longstanding challenge in robot-learning manipulation. Learning based approaches have shown promise in producing robust policies, but require heavy supervision to efficiently learn precise control, especially from visual inputs. We propose a novel self-supervision technique that uses time-reversal to learn goals and provide a high level plan to reach them. In particular, we introduce the time-reversal model (TRM), a self-supervised model which explores outward from a set of goal states and learns to predict these trajectories in reverse. This provides a high level plan towards goals, allowing us to learn complex manipulation tasks with no demonstrations or exploration at test time. We test our method on the domain of assembly, specifically the mating of tetris-style block pairs. Using our method operating atop visual model predictive control, we are able to assemble tetris blocks on a physical robot using only uncalibrated RGB camera input, and generalize to unseen block pairs.
Click here to read more.