Bolei Chen, Yongzheng Cui, Haonan Yang, Ping Zhong, Yu Sheng, Miao Li, Jianxin Wang
Object Target Search (OTS) tasks require robots to localize and navigate to objects specified by semantic labels, (e.g., find a fire extinguisher). Many existing OTS methods strongly rely on the semantic co occurrence relations among objects to imagine and localize the object targets seen in the learning phase. However, simplistic domestic and chaotic rescue scenarios often fail to provide rich semantics even require the robot to navigate to novel object instances. In addition, most of the existing OTS methods agree on a fixed Field of View (FoV) setting relative to the robot base, thus making it challenging to seek obscured objects (e.g., a shoe under the bed). To alleviate the above problems, we propose a semantic-heuristic OTS algorithm by promptly updating an Object Belief Field (OBF) for localizing object targets and designing a utility function for balancing exploration and exploitation. In particular, our method employs a flexible vision and allows the manipulator to translate and rotate the robot’s FoV to look around or even inspect hidden corners obscured by obstacles. Sufficient comparative and ablation studies validate that our method significantly improves the success rates and SPL metrics relative to the baselines. Furthermore, real-world experiments demonstrate our method can find novel objects without requiring rich scene priors.
This repository contains code for SHOTS.
This is a demo video of the experiments for the paper titled “SHOTS: Semantic-Heuristic Object Target Search with Object Belief Field and Flexible Vision”.
Clone
git clone https://github.com/BoLeiChen/SHOTS.git
Compile
cd VM-SHOTS
catkin build
Configure the conda environment
cd ..
git clone https://github.com/ultralytics/ultralytics.git
conda create -n SHOTS python=3.7
conda activate SHOTS
cd ultralytics
pip install -r requirements.txt -i https://mirrors.bfsu.edu.cn/pypi/web/simple/
pip install ultralytics
Sumilation enviromnent
source ./devel/setup.bash
roslaunch husky_ur5 husky_empty_world.launch
Local planner
source ./devel/setup.bash
roslaunch vehicle_simulator system_real_robot.launch
NBV Planner
source ./devel/setup.bash
roslaunch fkie_nbv_planner run_planner.launch
Load CLIP(Run in SHOTS environment)
source activate SHOTS
cd ./src/fkie-nbv-planner/scripts
python semantic_cloud.py
Experimental diagram
We simulate the search for casualties scenario in a 10 x 10 x 2 (
We deploy VM-SHOTS to the Husky-UR5 robotics platform to conduct experiments in a 8 × 15 × 6 (
This repository is released under the () license. See LICENSE for additional details.
Code acknowledgements:
src/fkie-nbv-planner/scripts/semantic_cloud.py
modified from Chefer et al.'s codebase, which was in turn modified from the original CLIP codebase.
Coming soon ...