How dongchenyu165/dongdoctoralresearch works
Project type and purpose
• This is a C++-based research tool (with Python bindings) for evaluating robotic grasp poses in a bi-arm knife-cutting task.
• It implements a multi-stage pipeline that, given a knife trajectory and food surface mesh, generates candidate grasp positions, filters them geometrically, computes force and position scores, aggregates them, and outputs the best grasp.
File structure
• Top-level directories correspond to pipeline stages:
• 1_PrepareData
• 2_FilterByGeo
• 3_KnifeForceCalculator
• 4_ForceScore
• 4_PositionScore
• Supporting modules:
• SearchSpaceGenerator – produces the initial candidate set
• ScoreCalculator – orchestrates per-stage scoring and final aggregation
• Shared code:
• DataTypes – definitions for trajectory nodes, point sets, etc.
• Utilities – numerical routines (Eigen helper), EMST (CGAL), PCL helpers, JSON serialization, logging (spdlog)
• Entrypoints:
• main.cpp / main_grasping_pc.cpp – command-line applications
• pybind11 submodule – Python bindings
• Build and test:
• CMakeLists.txt at root and per-module
• Tests/ – unit tests for each module
README for insights
• The pipeline has six logical stages:
1. Input knife trajectory → extract pose Tj
2. PrepareData: compute contact surface Ωc and gripping surface Ωg, initial search space P_init
4. KnifeForceCalculator: compute forces from the knife on the food
5. ForceScore: compute finger force distribution score (uses CUDA optionally)
6. PositionScore: compute fingertip positional redundancy score
7. Aggregate scores over trajectory → pick best grasp
• Key dependencies: Eigen, nlohmann/json, spdlog, cppitertools, CGAL, PCL, VTK, pybind11
Elements
• Rectangles for modules (SearchSpaceGenerator, PrepareData, …)
• A “Shared Utilities” rectangle beneath them
• Utilities in another
• External dependencies in a third