A preprocessing-based planning framework for utilizing contacts in high-precision insertion tasks

TL;DR

This work tackles high-precision insertion under pose uncertainty by framing localization via contact observations as a finite-set POMDP. It introduces a preprocessing-based framework that builds a database of policies and an experience-driven solver, E-RTDP-Bel, to accelerate planning across related problems. The approach achieves substantial preprocessing speedups and robust performance, demonstrated on real plug insertion with port pose uncertainty and in simulated pipe assembly with 4D pose uncertainty. The results suggest that combining binary tactile observations with experience-based planning can enable reliable, time-critical high-precision manipulation in semi-structured environments.

Abstract

In manipulation tasks like plug insertion or assembly that have low tolerance to errors in pose estimation (errors of the order of 2mm can cause task failure), the utilization of touch/contact modality can aid in accurately localizing the object of interest. Motivated by this, in this work we model high-precision insertion tasks as planning problems under pose uncertainty, where we effectively utilize the occurrence of contacts (or the lack thereof) as observations to reduce uncertainty and reliably complete the task. We present a preprocessing-based planning framework for high-precision insertion in repetitive and time-critical settings, where the set of initial pose distributions (identified by a perception system) is finite. The finite set allows us to enumerate the possible planning problems that can be encountered online and preprocess a database of policies. Due to the computational complexity of constructing this database, we propose a general experience-based POMDP solver, E-RTDP-Bel, that uses the solutions of similar planning problems as experience to speed up planning queries and use it to efficiently construct the database. We show that the developed algorithm speeds up database creation by over a factor of 100, making the process computationally tractable. We demonstrate the effectiveness of the proposed framework in a real-world plug insertion task in the presence of port position uncertainty and a pipe assembly task in simulation in the presence of pipe pose uncertainty.

Figures (6)

• Figure 1: Experimental setup for plug insertion using a UR10e manipulator.
• Figure 2: Belief tree for the active localization using contacts problem. The robot uses the occurrence of contacts (or the lack thereof) to reduce uncertainty, i.e., the hypothesis pose set.
• Figure 3: left) $b_{start}$, $\mathcal{B}^E$, and $\mathcal{G}$ are highlighted in yellow, maroon, and green respectively. The instantaneous transitions/jumps relevant for computing $\text{heur}^E$ for $b_{start}$ are indicated with dashed arrows. Under a sufficiently high $\varepsilon$, the experience heuristic prefers the policy highlighted in red, encouraging the search to move towards $\mathcal{B}^E$, and the value of this policy is used as $\text{heur}^E(b_{start})$. middle and right) Contrast the region of the belief space $\mathcal{B}$ explored by RTDP-Bel to converge to a solution (highlighted in blue) when using $\text{heur}^E$ (under high $\varepsilon$) vs $\text{heur}$, respectively.
• Figure 4: left) Getting from the start belief of a more complex problem $p'$ (yellow) to the solution of a simpler problem $p$, i.e., $\mathcal{B}^E_{naive}$ (maroon) can be non-trivial. right) Rolling out the solution of the $p$ from $p'_{start}$ makes the experience directly accessible from $p'_{start}$.
• Figure 5: A run of the proposed framework on the plug insertion task in the real world (top row) and the pipe assembly task in simulation (bottom row). We observe the robot making contacts at different locations to localize the object of interest and completing the insertion task upon localization. In the pipe assembly case, the construction of complex structures can be modeled as a sequence of insertions.