Parameter tuning versus early error detection for improving assembly performance

Determine whether optimizing the hand-coded threshold parameters governing the force- and torque-based stopping criteria in the spiral-based search (InsertPartSpiral) and tilt-based insertion (InsertTilt) algorithms yields meaningful improvements in assembly success rates compared to developing improved methods for early error detection and correction during autonomous industrial assembly.

Background

The system uses simple, hand-coded force and torque thresholds as stopping criteria for core assembly primitives (spiral search and tilt insertion). Despite demonstrating baseline success rates, the authors question the value of further tuning these parameters.

They explicitly raise uncertainty about whether effort should focus on parameter optimization or on advancing online error detection and correction methods that could identify and recover from failures earlier during execution.

References

Although the success rates in Table \ref{successTable} suggests that there is significant room for improvement, it is unclear whether finding better parameters would actually help much, or if we should not rather investigate better methods for early error detection and correction.

— Autonomous Industrial Assembly using Force, Torque, and RGB-D sensing  (2002.02580 - Watson et al., 2020) in Section 7 (Discussion)

While we showed the execution of the plans on real robots, closed loop execution of plans is left as future work. The controllers currently absorb some inaccuracies, but a replanning approach is required to deal with larger execution failures.

— WRAP: Fixtureless Wrench-aware Multi-Robot Assembly Planning  (2609.29407 - Hartmann et al., 24 Sep 2026) in Section Discussion {content} Limitations