Safe and Efficient Robot Manipulation: Task-Oriented Environment Modeling and Object Pose Estimation

Ying Zhang,Guohui Tian,Xuyang Shao

DOI: https://doi.org/10.1109/tim.2021.3071222

IF: 5.6

2021-01-01

IEEE Transactions on Instrumentation and Measurement

Abstract:Mobile manipulation is essential for robots to accomplish everyday household chores such as set the table. In order to successfully perform manipulation tasks, a 3-D representation of the environment and the pose of the target object are needed. In this article, we propose a systematic solution for safe and efficient robot manipulation. Innovatively, a task-oriented environment modeling strategy is presented for collision-free navigation and motion planning, which integrates the grid-based 2-D map and local real-time octree-based 3-D representation. Furthermore, we introduce a task-oriented object pose estimation approach based on the fiducial marker and ontology technology. In particular, a property-driven target object inference and pose estimation algorithm is designed, which allows the mobile robot to implement the task-oriented object manipulation. The proposed solution is evaluated through real-world experiments, where memory usage, computation time, collision checking, and object pose estimation are extensively investigated. Also, application scenarios are presented to validate the effectiveness and efficiency of our proposal.

engineering, electrical & electronic,instruments & instrumentation

What problem does this paper attempt to address?

The problem that this paper attempts to solve is: in order to enable robots to complete operational tasks such as daily housework (such as setting the table) safely and efficiently, it is necessary to establish a 3D representation of the environment and pose estimation of target objects. Specifically, the paper proposes a systematic solution to achieve safe and efficient robot operations, including the following two main aspects:

1. **Task - oriented environmental modeling**:
- A method combining 2D grid maps and local real - time 3D representations is proposed for collision - free navigation and motion planning. This method combines grid - based 2D maps with octree - based local real - time 3D representations, thereby reducing the need for complete 3D modeling of the entire environment, reducing memory consumption and improving computational efficiency.

2. **Task - oriented target object pose estimation**:
- A task - oriented target object pose estimation method based on fiducial markers and ontology technology is introduced. Through attribute - driven target object reasoning and pose estimation algorithms, mobile robots can accurately identify and operate target objects according to task requirements.

### Formula summary

The key formulas involved in the paper are as follows:

- **Distance measurement formula**:
\[
D_r^\tau < \xi L_r
\]
where \( D_r^\tau \) represents the distance from the target object to the robot, \( \xi \) is the safety factor, and \( L_r \) is the length of the robotic arm.

- **Occupancy probability update formula**:
\[
P(v_i | \tilde{o}_{1:k})=\left[1+\frac{P(v_i)}{1 - P(v_i)}\cdot\frac{P(v_i | \tilde{o}_k)}{1 - P(v_i | \tilde{o}_k)}\cdot\frac{P(v_i | \tilde{o}_{1:k - 1})}{1 - P(v_i | \tilde{o}_{1:k - 1})}\right]^{-1}
\]
where \( P(v_i) \) and \( P(v_i | \tilde{o}_{1:k - 1}) \) represent the prior probability and the previous estimate respectively.

- **Log - odds representation**:
\[
Lg(v_i | \tilde{o}_{1:k}) = Lg(v_i | \tilde{o}_k)+Lg(v_i | \tilde{o}_{1:k - 1})
\]

- **Final occupancy probability**:
\[
P(v_i)=\frac{e^{Lg(v_i | \tilde{o}_{1:k})}}{1 + e^{Lg(v_i | \tilde{o}_{1:k})}}
\]

- **Voxel state determination**:
\[
S(v_i)=
\begin{cases}
\text{occupied}, & \text{if } P(v_i)>P_\varsigma^o\\
\text{free}, & \text{if } P(v_i)<P_\varsigma^f\\
\text{unknown}, & \text{if } P_\varsigma^f\leq P(v_i)\leq P_\varsigma^o
\end{cases}
\]

### Summary

The main contribution of this paper is to propose an environmental modeling strategy that combines 2D OGM and local real - time 3D representations, as well as a task - oriented target object pose estimation method based on markers and ontology technology, thereby achieving safe and efficient robot operations. Through experimental verification, this method performs well in terms of memory consumption, computation time, collision detection, object detection, and pose estimation.