Latent Representations for Visual Proprioception in Inexpensive Robots
This addresses the challenge of enabling accurate manipulation in low-cost robots operating in unstructured environments, though it is incremental as it applies existing methods to a new setting.
The paper tackled the problem of estimating joint positions for inexpensive robots lacking precise proprioception by using a single external camera image, achieving a regression accuracy of 0.95 mm in experiments on a 6-DoF robot.
Robotic manipulation requires explicit or implicit knowledge of the robot's joint positions. Precise proprioception is standard in high-quality industrial robots but is often unavailable in inexpensive robots operating in unstructured environments. In this paper, we ask: to what extent can a fast, single-pass regression architecture perform visual proprioception from a single external camera image, available even in the simplest manipulation settings? We explore several latent representations, including CNNs, VAEs, ViTs, and bags of uncalibrated fiducial markers, using fine-tuning techniques adapted to the limited data available. We evaluate the achievable accuracy through experiments on an inexpensive 6-DoF robot.