Online Whole-body Motion Planning for Quadrotor using Multi-resolution Search
Y. Ren, S. Liang, F. Zhu, G. Lu and F. Zhang, βOnline Whole-Body Motion Planning for Quadrotor using Multi-Resolution Search,β 2023 IEEE International Conference on Robotics and Automation (ICRA), London, United Kingdom, 2023, pp. 1594-1600, doi: 10.1109/ICRA48891.2023.10160767.
Liu et al. κ³Ό Han et al. μ΄ κ°μ§κ³ μλ λ¬Έμ λ₯Ό μΌλΆ ν΄κ²°νλ©΄μ λͺ¨λ₯΄λ μ§λμ λν μ΅μ κ²½λ‘ μμ±μ λν λ Όλ¬ΈμΌλ‘, λλ€μλ GCOPTER λ Όλ¬Έμ κΈ°λ°μΌλ‘ νλ νλ μ μν¬λ₯Ό μ μνμλ€.
μ£Όμ contribution
- μ’μ μ§μμ full pose (=) planningμ νκ³ μ΄μΈμ μ§μμ position planningμ νλλ‘ νλ hierachical framework λ₯Ό μ μ
- λΌμ΄λ€λ₯Ό μ΄μ©ν΄ λͺ¨λ₯΄λ μ§νμμ λΉνμ΄ κ°λ₯νκ³ μ μν νλ μ μν¬λ₯Ό ν΅ν΄ μ¨λ³΄λλ‘ μ»΄ν¨ν μ΄ κ°λ₯ν¨
λ¬Έμ μ
- μ€ν μμ€λ‘ λ΄μ©μ 곡κ°νμ§ μμλ€
- κ·Έλ¦¬κ³ μμ νμμ²΄λ‘ λͺ¨λΈλ§ νλ κ²μμ μ΄μκ° λ°μν μ μμ
- μλ‘ μ μν μ¬λ¬ λ°©μμμ κ°μ μ μ΄ μμ μ μμ
- Parallel multi-resolution search: λ 볡μ‘ν νκ²½ νΉμ νκ³Ό λμ 곡κ°μ΄ κ°μ΄ μλ κ²½λ‘μμ λΉν¨μ¨μ μΈ κ²½λ‘ κ³ν κ°λ₯
- Seed generation method: κΈ°μ‘΄ polyhedron μμ± λ°©λ²μμ overlapμ΄ μκΈ°λ κ²μ κ°μ νκ³ μ λμ ν λ°©μμΈλ°, λ°©μμ΄ λͺ λ£νμ§ μμ.(κ·Έλ¬λ μ½λκ° κ³΅κ°λμ§ μμμ μ΄λ€ μμΌλ‘ μ€νμμ μ μ©νμλλ° λΆλΆλͺ ν¨)
- μ κ²½λ‘ κ°μ μ°κ²°
Introduction
planning λ Όλ¬Έλ€κ³Ό planning λ Όλ¬Έλ€μ μ μνκ³ λ¨μ λ€μ μ§μ νλ λΆλΆλ€λ§ μμ½ν΄μ μ 리ν¨.
Simply ignore the shape and orientation of the drone
- Jesus Tordesillas, Brett T Lopez, Michael Everett, and Jonathan P How. Faster: Fast and safe trajectory planner for navigation in unknown environments. IEEE Transactions on Robotics, 38(2):922β938, 2021.
- Xin Zhou, Zhepei Wang, Hongkai Ye, Chao Xu, and Fei Gao. EGO-Planner: An ESDF-free gradient-based local planner for quadrotors. IEEE Robotics and Automation Letters, 6(2):478β 485, 2021.
- Yunfan Ren, Fangcheng Zhu, Wenyi Liu, Zhepei Wang, Yi Lin, Fei Gao, and Fu Zhang. Bubble planner: Planning high-speed smooth quadrotor trajectories using receding corridors. arXiv preprint arXiv:2202.12177, (Accepted by 2022 IROS), 2022.
- Boyu Zhou, Fei Gao, Jie Pan, and Shaojie Shen. Robust real-time UAV replanning using guided gradient-based optimization and topological paths. In 2020 IEEE International Conference on Robotics and Automation, pages 1208β1214. IEEE, 2020.
- Sikang Liu, Nikolay Atanasov, Kartik Mohta, and Vijay Ku-mar. Search-based motion planning for quadrotors using linear quadratic minimum time control. In 2017 IEEE/RSJ international conference on intelligent robots and systems (IROS), pages 2872β2879. IEEE, 2017.
μ λ Όλ¬Έλ€μ λλ‘ λͺ¨λΈλ§μ κ΅¬λ‘ νκ±°λ orientation μ 무μνμ¬ μμ νλλνλ μ ν μ°κ΅¬λ€μ΄λ€.
μ μλ €μ§ Faster (T-RO λ Όλ¬Έ)κ³Ό EGO-Plannerμ Bubble Plannerλ 보μ΄κ³ μμ μ 리νλ Search-based planning using LQMTC λ Όλ¬Έλ 보μΈλ€.
μ΄λ¬ν λ°©μμ μμ μ¬λ¬μ°¨λ‘ λ€λ€λ λλ‘ μμΈλ₯Ό κ³ λ €νμ§ μμ 보μμ μΌλ‘ λΉννκ³ dynamically feasible νλ€κ³ 보기 μ΄λ ΅λ€.
planning λ Όλ¬Έλ€μ ν° λ¬Έμ μ€ νλλ unknown and unstructured νκ²½μμ κ²½λ‘ κ³ννμ§ μλλ€λ κ²μ΄λ€.
. but assumptions on the environment (small gaps)
- Giuseppe Loianno, Chris Brunner, Gary McGrath, and Vijay Kumar. Estimation, control, and planning for aggressive flight with a small quadrotor with a single camera and imu. IEEE Robotics and Automation Letters, 2(2):404β411, 2016.
- Toru Hirata and Makoto Kumon. Optimal path planning method with attitude constraints for quadrotor helicopters. In Proceedings of the 2014 International Conference on Advanced Mechatronic Systems, pages 377β381. IEEE, 2014.
- Jiarong Lin, Luqi Wang, Fei Gao, Shaojie Shen, and Fu Zhang. Flying through a narrow gap using neural network: an end-to-end planning and control approach. In 2019 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pages 3526β3533. IEEE, 2019.
- Shaohui Yang, Botao He, Zhepei Wang, Chao Xu, and Fei Gao. Whole-body real-time motion planning for multicopters. In 2021 IEEE International Conference on Robotics and Automation (ICRA), pages 9197β9203. IEEE, 2021.
- Zhepei Wang, Xin Zhou, Chao Xu, and Fei Gao. Geometrically constrained trajectory optimization for multicopters. IEEE Transactions on Robotics, 2022.
μ λ Όλ¬Έλ€μ μ§λκ°μΌ νλ νμ λν΄ λ―Έλ¦¬ μκ³ μλ κ²μ μ μ λ‘ κ²½λ‘ κ³νμ΄ μ΄λ£¨μ΄ μ Έμ μ€μ νκ²½μμ μ¬μ©νκΈ°μλ μ΄λ ΅κ±°λ μ μ μ§λκ° νμνλ€.
Han et al. λ Όλ¬Έμμ μΈκΈλμλ―μ΄ [6]-[8] μ νμ μ§λλ κ²μλ§ μ΄μ μ λ§μ΄ λ§μΆμκ³ , [9] λ μ΄μ Post μμ λ€λ£¨μκ³ [10] μ GCOPTER λ Όλ¬Έμ΄λ€.
λ λ€λ₯Έ κΈ°μ‘΄ Planning μ°κ΅¬μ λ¬Έμ μ μΌλ‘ κΌ½μ κ²μ μ°μ° ν¨μ¨μ±μ΄λ€. μλ§λ λ³Έ μ°κ΅¬μ contributionμ κ°μ‘°νκ³ μ λ£μ κ²μΌλ‘ 보μ΄λλ°, μλ μ°κ΅¬λ€μ κΌ½μΌλ©° μλ°± μμ μ μ΄μ μ°μ°μ΄ κ±Έλ¦°λ€κ³ νλ€.
Recent online planning researchs
- Zhichao Han, Zhepei Wang, Neng Pan, Yi Lin, Chao Xu, and Fei Gao. Fast-racing: An open-source strong baseline for se(3) planning in autonomous drone racing. IEEE Robotics and Automation Letters, 6(4):8631β8638, 2021.
- Sikang Liu, Kartik Mohta, Nikolay Atanasov, and Vijay Kumar. Search-based motion planning for aggressive flight in se (3). IEEE Robotics and Automation Letters, 3(3):2439β2446, 2018.
Sikang λ Όλ¬Έ κ³Ό κ°μ΄ λ¬Άμ¬μ κΉμΈ λλμ΄ μμ§λ§, Han et al. μ κ²½μ°μλ μ΄μ κ²½λ‘ κ³νμ νμλ‘ νκ³ μ¬μ€ Yang et al. λ Όλ¬Έμ GPU λ³λ ¬ μ»΄ν¨ν μΌλ‘ μ΅μ ν ν κ²μ΄λ―λ‘ μ¬μ ν κ½€ λ§μ μκ°μ΄ μμλλ€κ³ λ³Ό μ μλ€.
λ°λΌμ λ³Έ μ°κ΅¬μμλ Multi-resolution search method λ₯Ό ν΅ν΄ λμ μ§μμμλ planning μ μννκ³ online planning μ€ λ°κ²¬λλ μ’μ νμ λν΄μ planning μ μννμ¬ μ λ¬Έμ λ€μ ν΄κ²°νκ³ μ νλ€.
Related Works
μμμλ μΆ©λΆν μ ν μ°κ΅¬μ λν λ Όμκ° μ΄λ£¨μ΄μ Έμ νμν λΆλΆ μμ£Όλ‘λ§ μμ½.
μμμ [1]-[3] (Faster, EGO, Bubble-planner) κ³Ό μλ λ λ Όλ¬Έλ€μ 기체λ₯Ό κ΅¬λ‘ λͺ¨λΈλ§νμ¬ configuration space λ₯Ό μ½κ² ꡬμ±ν μ μμκ³ μ₯μ λ¬Ό ννΌμ λν 보μμ μΈ κ°μ μ΄ μ μ λ planning μ μννμλ€.
Quote
- Boyu Zhou, Fei Gao, Luqi Wang, Chuhao Liu, and Shaojie Shen. Robust and successful quadrotor trajectory generation for fast autonomous flight. IEEE Robotics and Automation Letters, 4(4):3529β3536, 2019.
- Fanze Kong, Wei Xu, Yixi Cai, and Fu Zhang. Avoiding dynamic small obstacles with onboard sensing and computation on aerial robots. IEEE Robotics and Automation Letters, 6(4):7869β7876, 2021.
κ·Έλ¦¬κ³ [9]-[11] μΈ Yang et al., GCOPTER, Han et al. λ Όλ¬Έλ€μ΄ optimization-based λ°©μμ μ΅μ λ Όλ¬Έλ€μΈλ°, λ¬Έμ μ μΌλ‘ κΌ½μ κ²λ€μ μλμ κ°λ€.
μ°μ μ΅μ ν μ μ μνλλ μ μ κ²½λ‘ νμμ ν¬ν¨νλ©΄ μλ°± κ° μμλλ€λ μ κ³Ό, collision-free λ₯Ό μν΄ μ¬μ©λλ flight corridor λ°©μμ΄ λλλ‘ infeasible νμ¬ μ’μ νμ μ§λλ κ²μ μ€ν¨νλ κ²½μ°κ° μλ€λ κ²μ΄λ€.
λ³Έ μ°κ΅¬λ GCOPTER λ Όλ¬Έμ λλΆλΆ λ°©μμ μ¬μ©νλ μ planning μ νΌν©νκ³ flight corridor λ₯Ό μΌλΆ μμ νμ¬ μ£Όμ΄μ§ μ§λ μμ΄, μ€μκ°μΌλ‘ μ’μ νμ λμ μ±κ³΅λ₯ κ³Ό μ μ ν μ»΄ν¨ν μκ° μΌλ‘ λΉνν λ‘ νμλ€.
Preliminaries
A. System Modeling and Polynomial Trajectory
μ¬κΈ°λ λ§μ΄ λ€λ€λ differentially flat system μ λν λ΄μ©μ΄λ€. λ€λ₯Έ λ Όλ¬Έλ€μμλ yaw angle trajectory λ₯Ό 0μΌλ‘ λλ κ²½μ°κ° λ§μλλ° μ¬κΈ°μλ μ§ν λ°©ν₯μ λ°λΌλ³Ό μ μλλ‘ μ λ²μ λ°©ν₯μΌλ‘ μ€μ νμλ€.
κ·Έλ¦¬κ³ kinodynamic constraints λν μ΄λ₯Ό μ΄μ©ν΄ μλ, κ°μλ, jerk μ μ/ννμΌλ‘ λμλ€. GCOPTER λ Όλ¬Έ λ°©μμ μ°¨μ©νμ¬ piece-wise polynomials λ‘ κ²½λ‘λ₯Ό νννμ¬, λ€νμ κ³μμ μκ° λ²‘ν°λ‘ μ μνμλ€.

μ¬κΈ°μ μ΄μ§ ν·κ°λ¦¬λ λΆλΆμ μΌλ‘ μ΄μ λ Όλ¬Έλ€κ³Ό ν¬κΈ°κ° λμΌνλ°, λ‘ λμ΄ jerk κΉμ§ μ°μμ μΌ μ μλλ‘ νμλ€κ³ νλ€.
μ΄μ λ Όλ¬Έμμλ μΌλ‘ λκ³ snap κΉμ§ μ°μμ μ΄λΌκ³ νννμλ€. (μλ μ΄λ―Έμ§)


보λ€μνΌ boundary condition μ΄λ―λ‘ μ°¨μλ κΉμ§μ¬μ λ 벑ν°λ€(λ€νμ κ³μ, μκ° κΈ°μ 벑ν°) ν¬κΈ°κ° κ°λ€.
Question
- κ°μ λ°λ₯Έ λ€νμ κΆ€μ μ°μμ±
- μ°μ μ£Όμ΄μ§λ initial and final state μ ν¬κΈ°μ λ°λΌ λ€νμμ κ³μμ μ°¨μκ° μ ν΄μ§λ€.
- λ³Έ λ Όλ¬Έμμλ λ‘ initial and final state λ₯Ό μ£Όλ―λ‘ κ° λ§κ³ 7μ°¨ λ€νμμ΄ λ§λ€μ΄μ§λ©° μ΄λ jerk κΉμ§ μ°μμ±μ νμλ‘ νλ€. λν, control inputμΌλ‘ snapμ μ¬μ©νλ κ²μΌλ‘ 보μΈλ€.(λ Όλ¬Έ μ€λͺ λͺ¨λ λ§μ)
- μ΄μ λ Όλ¬Έμμλ μΌλ‘ μ¬μ©νκ³ jerk λ₯Ό control input μΌλ‘ μ¬μ©νλ€. κ·Έλ λ€λ©΄ snap is always continuous on the whole trajectory κ° νΌλλλ€.
- jerkλ₯Ό control input μΌλ‘ νλ©΄ μμκ³Ό λ§μ§λ§ state λ κ°μλκΉμ§ μ 보λ₯Ό μ€λ€λ κ²μ΄λ€. λμ μ€κ° waypointμ μμΉκ° μ£Όμ΄μ§λ―λ‘ Optimality Condition μ μν΄ 4μ°¨μΈ snap κΉμ§ μ°μμ μ΄λ€.
- κ·Έλ λ€λ©΄ λ³Έ λ Όλ¬Έμμλ κ° piece-wise polynomials κ°μ μ°μμ μν΄ μΆκ°μ μΈ μ‘°κ±΄μ΄ νμνμ§ μλ?
B. Safety Constraints
collision-free ν κ²½λ‘λ₯Ό μν΄ safe flight corridor (SFC) λ₯Ό μ¬μ©νλ€.
κ° κ²½λ‘ κ³ν λ°©μμ λ°λΌ λ€λ₯Έ λͺ¨λΈλ§μ μ¬μ©νλ€. μμλ ν¨μ¨μ±μ μν΄ κ°μ₯ κΈ΄ μΆμ κΈΈμ΄ μ λ°μ§λ¦μ κ°μ§λ κ΅¬λ‘ νννλ€.
κ·Έλ¦¬κ³ λλ‘ λͺ¨λΈμ μ μΌλ‘ νννλ λμ μ₯μ λ¬Όλ€μ λ§νΌ νμ₯νμ¬ μκ°νλ©΄ (the configuration space by inflating all obstacle points with ) μλμ κ°μ μμμΌλ‘ SFC λ₯Ό ννν μ μλ€.
κ·Έλ¦¬κ³ μμλ νμ체λ₯Ό λμ νμ¬ Liu et al. κ³Ό κ°μ΄ μ¬μ©ν μ μλ€. λ€λ§, Liu et al. μμλ νμ체μ μ₯μ λ¬Ό μ κ³Όμ κ΅μ μ μνλ§μ ν΅ν΄ νλ¨νμμ§λ§, μ¬κΈ°μλ -representation λ°©μμΌλ‘ νννμλ€. λ²μ§Έ polyhedron μμ μλ λ²μ§Έ κΆ€μ μ λν΄ SFC μμμ μλμ κ°λ€.
μ΄ λ λ rotation matrix μ΄κ³ λ‘ νμ체 λͺ¨μμ ννν κ²μ΄λ€.
λΆλΆμ unit sphereμ λν΄ νμ체μ ν¬κΈ°λ§νΌ μλμ polyhedronμ΄ λ¨μ΄μ Έ μλμ§λ‘ μ΄ν΄νμλ€. μ¦ μλμ CoM λμ νμ체 ν¬κΈ°λ§νΌ μΆ©λνμ§ μλμ§λ₯Ό νλ¨νλ κ²μ΄λ€.
Note
- Yang et al. μμλ Liu et al. μμ νμμ²΄λ‘ λͺ¨λΈλ§ ν κ²μ λΉννλ©΄μ μ₯μΆμ κΈΈμ΄λ‘λ§ νλ©΄ CoM μ΄ μ€μμ΄ μλ κ²½μ°μ λ°μν μ μλ μ μ μ§μ νλ€.
- κ·Έλμ Yang et al. κ³Ό Han et al. μμλ vertex λ₯Ό κ°μ§κ³ λͺ¨λΈλ§ νμ¬ μ§μ‘면체λ λ€λ©΄μ²΄λ‘ λͺ¨λΈλ§νμλ€.
- λ³Έ λ Όλ¬Έμμλ κ·ΈλΌμλ νμ체λ₯Ό μ¬μ©ν μ΄μ κ° λ¬΄μμΌκΉ? μ°μ° μλλ₯Ό μ€μ΄κΈ° μν¨μ΄λ νμ체 λ§μΌλ‘λ μΆ©λΆν μ₯μ λ¬Ό ννΌκ° μ λμ΄μμΈκ°?
C. Trajectory Generation
planning μ μ£Όμ΄μ§ start state μ goal state λ₯Ό κ°μ§κ³ collision-free νκ³ kinodynamic constraints λ₯Ό λ§μ‘±νλ κ²½λ‘λ₯Ό μμ±νλ€.
RILS
- S. Liu et al., βPlanning dynamically feasible trajectories for quadrotors using safe flight corridors in 3-D complex environments,β IEEE Robot. Automat. Lett., vol. 2, no. 3, pp. 1688β1695, Jul. 2017.
μ΄μ Fast-racing(Han et al.) μμλ μΈκΈλ RILS λ°©μμΌλ‘ SFC λ₯Ό μ°μμ μΌλ‘ λ§λ€κ³ GCOPTER λ Όλ¬Έμ MINCO κ²½λ‘ μ΅μ νλ‘ μμ±νλ€.
κ·Έλ¦¬κ³ κ²½λ‘ μμ±μ λ§μ°¬κ°μ§λ‘ GCOPTER λ Όλ¬Έ λ°©μμ μ¬μ©νλ€.
Planner

A. Colliding Segments Extraction
μ°μ μ₯μ λ¬Όμ κ³ λ €νμ§ μκ³ Linear Quadratic Minimum Time λ¬Έμ λ₯Ό νΈλ λ°©μμΌλ‘ global start state μ global goal state κ°μ κ²½λ‘λ₯Ό μμ±νλ€.
Note
- λ Όλ¬Έ notation μμ global start state μ μ΄μ μ±ν° goal state μ ννμ΄ κ²ΉμΉλ€.
LQMT
- Mark W Mueller, Markus Hehn, and Raffaello DβAndrea. A computationally efficient motion primitive for quadrocopter trajectory generation. IEEE transactions on robotics, 31(6):1294β 1310, 2015.
λ€λ₯Έ λ Όλ¬Έλ€μμ μ§μ ν μ μλ μ μ κ²½λ‘ μμ±μ λ‘ νμ¬ dynamics κ° κ³ λ €λμ§ μμ λΆλΆμ ν΄κ²°ν κ²μΌλ‘ 보μΈλ€.
κ·Έ νμλ μμ±λ κ²½λ‘μμ λ°μνλ μΆ©λ μ§μ μ μ°ΎμλΈλ€. μμΈν λ΄μ©μ Fig.3 μ (a) λ₯Ό 보μ.
Question
- μμ μ°Ύμλ΄λ λ°©μ. μ΄λ€ μ§μ μμ κ° μ§μ μ λν μ 보λ₯Ό κ²°μ μ§λμ§?
B. Parallel Multi-resolution Search
λ³Έ λ Όλ¬Έμμ μ μνλ multi-resolution search λ μ°μ low-resolution map (LRM) κ³Ό high-resolution map (HRM) μΌλ‘ ꡬμ±λλ€. LRM μ 기체μ μ₯μΆ κΈΈμ΄ λ‘ λ§λ€μ΄μ§κ³ HRM μ λ¨μΆ κΈΈμ΄ λ‘ λ§λ€μ΄ μ§λ€.
μ΄λ₯Ό μν΄ νμμ²΄λ‘ μ μν κ²μΌλ‘ 보μΈλ€.
Note
- μ΄λ¬ν λ°©μμ λ¬Έμ μ μ μ΄μ μ μ§μ λ λλ‘ νμμ²΄κ° μ΄μμ μ΄μ§ μμ κ²½μ°μλ, μ’μ νμ ν΅κ³Ό μ¬λΆλ₯Ό 기체μ λ°λΌ νλ¦¬κ² νλ¨ν μ μλ€. (i.e. κ° κΈ°μ²΄μ κ°μ₯ μ§§μ λΆλΆλ³΄λ€ λ μ§§μ κ²½μ°)
λ°λΌμ LRM μ μμ λ°©μκ³Ό λμΌνλ―λ‘ Whole-body motion planning μ΄ νμνμ§ μκ³ , HRM μμ μ΅μ μ μ΄κ° νμνλ€κ³ λ³Ό μ μλ€.
κ° μ§λμ λν΄μ νμμ΄ μ΄λ£¨μ΄μ§κ³ κ°μ₯ λ¨Όμ νμμ΄ μλ£λ μ§λμ λν΄ κ²½λ‘ κ³νμ΄ μ΄λ£¨μ΄μ§λ€. (e.g. Terminal or Suspend sign)
planning μ μλνλλΌκ³ infeasible ν μ μμΌλ―λ‘ μ΄ λλ Suspend νλ κ²μ΄λ€.
C. SE(3) Trajectory Generation
μ΄μ μ΄μ μ±ν°μμ planning μ μλνλ κ²½μ°, μ΄λ₯Ό SE(3) sub-problem μΌλ‘ νννλ€.
μ΄μ λ‘λΆν° κΉμ§λ₯Ό μ°κ²°νλ SFC κ° νμνλ°, Yang et al., Han et al., RILS λ Όλ¬Έ λͺ¨λ λ λ Έλ(or state) κ°μ SFC λ₯Ό κ³§λ°λ‘ μμ±νκ³ μ΄λ λ corridor κ°μ overlapμ μΌκΈ°νκΈ°λ νλ€.

κ·Έλμ μμ±λ κ²½λ‘κ° infeasible ν΄μ§λ€κ³ νλ€.
Note
- μ infeasible ν΄μ§κ³ κΈ°μ‘΄ λ°©μμ μ΄λ€ λΆλΆ λλ¬ΈμΈμ§ μμΈν μκ°ν΄λ³΄κΈ°
λ³Έ λ Όλ¬Έμμλ μ΄ λ¬Έμ λ₯Ό ν΄κ²°νκΈ° μν΄ a simple seed generation method μ λμ νμλ€.

μ¬κΈ°μλ ꡬꡬμ μ μ€λͺ μ΄ λ€μ΄κ°μ κ·Έλ¦Όμ 보λ κ²μ΄ μ΄λ‘λ€.
κ·Έλ¦Ό (a) μμλ μ μ κ²½λ‘κ° μ₯μ λ¬Όμ λ§νμ§λ§ μμ μ’μ νμ μ§λ μ μκ² λ§λ€μ΄μ§ HRM μ ννν κ²μ΄λ€.
κ²½λ‘ λ νμμΌλ‘ λ§λ€μ΄μ‘κ³ κ²½λ‘μ μμ κ°μ₯ κ°κΉκ² μ₯μ λ¬Όλ‘ μΈμλ μ§μ (nearest occupied grid) λ₯Ό μ°Ύλλ€.
κ·Έλ¦¬κ³ λ°©ν₯μΌλ‘ νμνμ¬ μ²« λ²μ§Έλ‘ μ₯μ λ¬Όλ‘ μΈμλ μ§μ (the first occupied grid) λ₯Ό μ°Ύμ μ΄λ₯Ό μ λ£λλ€. μ΄ λ μ¬μ©λ λ μ λ£λλ€.
κ·Έλ¬λ©΄ μ΄μ μ’μ νμ κ°μ₯μ리(or μ₯μ λ¬Ό κ°μ₯μ리)λ‘ μκ°ν μ μλ μ νμ μ§λλ κ²½λ‘μ μ΄ μκΈ΄λ€. μ΄ μ νκ· λ°©ν₯μ λΌκ³ νκ³ μ μ€μ¬μ μ€μ¬μ λ‘ μ μνλ€.
μ κΈΈμ΄λ₯Ό λ°©ν₯ λ²‘ν° μ ν¬μν΄μ μ»μ΄μ§ κΈΈμ΄ λ‘ line seed λ₯Ό μ»λλ€. (Fig. 5. (c) μ°Έκ³ )
μ΄μ μ μ λμ μΌλ‘ polyhedron λ₯Ό λ§λ€κ³ λ₯Ό line seed λ‘ νμ¬ μ’μ νμ μ§λλ λ₯Ό λ§λ λ€.
μ΄λ¬ν λ°©μμΌλ‘ μμ±λ SFC μμ κ²½λ‘ μ΅μ νκ° safety & dynamic constraints λ₯Ό λ§μ‘±νλλ‘ μμ±λλ€λ©΄ SE(3) segment λ‘ λ§λ€μ΄μ§κ³ ,
κ·Έλ μ§ μμΌλ©΄ sub-problem μΌλ‘ λΆλ₯λμ΄ LRM μμ low-resolution search(LRS) κ° μΌμ΄λλ€.
D. R^3 Trajectory Generation
μ΄μ μΆ©λ μ§μ λ§λ€ μκΈΈ μλ μλ κ°μ SE(3) κ²½λ‘ μ‘°κ°λ€μ΄ μλ€κ³ νμ. μ΄ κ²½λ‘λ€μ start states μ goal states λ€λ‘ μ΄λ£¨μ΄μ Έ μλ€.
μ΄μ global start state μ 첫 SE(3) κΆ€μ , νΉμ μ°μμ μΈ SE(3) κΆ€μ μ΄λ λ§μ§λ§ SE(3) κΆ€μ κ³Ό global goal state λ€ μ¬μ΄μ LRS λ‘ μκΈ΄ κ²½λ‘ νΉμ μΆ©λμ΄ μλ μ μ κ²½λ‘ κ° μμ κ²μ΄λ€.
μ΄μ μλ λ°©μμΌλ‘ μ¬μ΄μ μλ state λ€μ΄ μμ κ²μ΄λ―λ‘ μ΄λ€μ κ°κ° start or goal state λ‘ νμ¬ κ²½λ‘ μ΅μ νλ₯Ό ν΄μ κ²½λ‘λ₯Ό λ§λ€μ΄ λΈλ€.

Experiments
A. Benchmark Comparison
μ ν μ°κ΅¬λ€κ³Ό λ€μνκ² λΉκ΅νμλ€. λ€λ§ λ³Έ μ°κ΅¬ μ체λ 곡κ°νμ§ μμμ μΆνμ λΉκ΅κ° μ΄λ €μμ‘λ€.
μμ½νμλ©΄ Han et al. λ³΄λ€ κΆ€μ μ΄ μ§§κ±°λ λΉ λ₯΄λ€κ³ 보긴 μ΄λ ΅μ§λ§, λΉμ·ν μμ€μ μ±λ₯μ λ΄λ λμ ν¨μ¬ λΉ λ₯Έ μλλ‘, μ£Όμ΄μ§ μ§λ μμ΄ μννλ€.

무μ보λ€λ Han et al. λ³΄λ€ μ±κ³΅λ₯ μ΄ μλ±ν λμΌλ©° search-based λ°©μμ Liu et al. μ΄ κ°μ₯ λμ§λ§ μ΄μͺ½μ μκ°μ΄ λ무 μ€λκ±Έλ €μ μ€μ¬μ©μλ μ΄λ ΅λ€.
B. Real-world Tests
μ€μ νκ²½μμμ ν μ€νΈκ° κ΅μ₯ν μ΅μ μ°κ΅¬λ€μ μ΄μ§ν©νμ¬ μννμλ€.
-live λ Fast-LIO2, Bubble Planner λ±μ μ¬μ©νμλ€.
Real-world tests
- Jiarong Lin and Fu Zhang. R3 live: A robust, real-time, rgbcolored, lidar-inertial-visual tightly-coupled state estimation and mapping package. In 2022 International Conference on Robotics and Automation (ICRA), pages 10672β10678, 2022.
- Jiarong Lin and Fu Zhang. R3 live++: A robust, real-time, radiance reconstruction package with a tightly-coupled lidarinertial-visual state estimator. arXiv preprint arXiv:2209.03666, 2022.
- Wei Xu, Yixi Cai, Dongjiao He, Jiarong Lin, and Fu Zhang. Fast-lio2: Fast direct lidar-inertial odometry. IEEE Transactions on Robotics, 2022.
- Fangcheng Zhu, Yunfan Ren, and Fu Zhang. Robust real-time lidar-inertial initialization. arXiv preprint arXiv:2202.11006, (Accepted by 2022 IROS), 2022.
- Guozheng Lu, Wei Xu, and Fu Zhang. Model predictive control for trajectory tracking on differentiable manifolds. arXiv preprint arXiv:2106.15233, 2021.
- Yunfan Ren, Fangcheng Zhu, Wenyi Liu, Zhepei Wang, Yi Lin, Fei Gao, and Fu Zhang. Bubble planner: Planning high-speed smooth quadrotor trajectories using receding corridors. arXiv preprint arXiv:2202.12177, (Accepted by 2022 IROS), 2022.
We use an on-manifold model predictive controller [24] to perform high-accuracy trajectory tracking. To realize online planning and cope with newly sensed obstacles during the flight, we adopt a distance-triggered receding horizon planning scheme from our previous work [3]
μ¬μ€ βTo realize online planning and cope with newly sensed obstacles during the flightβ λΆλΆμ΄ μ‘°κΈ μ΄ν΄κ°μ§ μμ§λ§ manifold μμμ MPC μ Bubble planner λ Όλ¬Έμ λ³Ό νμκ° μμ΄ λ³΄μΈλ€.

κ²°κ³Όμ λν μ¬μ§κ³Ό μμλ μ λμ μλ€. μΈμ κΉμλ κ²μ Fig. 13. μμ μκ°νλ₯Ό μ΄λ€ μμΌλ‘ νμλμ§ κΆκΈνλ€. Pointcloudκ° λ°ν¬λͺ Box ννλ‘ λμ€λ κ²λ 보기μ κ΅μ₯ν μ’μ보μ΄μ§λ§, λͺ¨λΈλ§ν λλ‘ μ μμ§μμ΄ νλ μ λ³λ‘ λͺ¨λΈ / νμμ²΄λ‘ μ ννλμ΄ μλ€.
Conclusion and Future work
λ³Έ λ Όλ¬Έμμ κΌ½μ νκ³μ μΌλ‘λ μ’μ νμ ν μͺ½λ§ μΈμνκΈ° λλ¬Έμ μ’μ νμ΄ κΈΈκ² λ μλ κ²½μ°μλ μ§λκ° μ μμ§λ§, μ΄λ₯Ό νμ ν μ μλ λ°©μμ κ²½λ‘ κ³νμ μλλ€.
μ΄λ₯Ό μν΄μλ lateral motion μΌλ‘ nonzero roll angle μ μ μ§ν μ μμ΄μΌ νλ€κ³ νλ€. μ΄λ₯Ό perception-aware planning μ ν΅ν΄ μ’μ νμ μ λλ‘ μΈμνμ¬μΌ νλ€κ³ νλ€.
Note
- μ‘°κΈ λ μκ°ν΄λ³΄λ©΄, λΌμ΄λ€μ μΌμ λ²μμ μν₯μ΄ μμ κ² κ°λ€. κ·Έλ¦¬κ³ μ’μ νμ΄λ whole-body motion planning μ΄ νμν μν©μ΄ μ’μ νμ λμ΄ μ¬νκ±°λ, μ°μμ μΌλ‘ μλ κ²½μ°λ μ΄λ€ μμΌλ‘ ν΄κ²°ν μ μμμ§ κΆκΈνλ€. (λλ€μλ unfeasible νμ¬ κ²½λ‘κ° μκΈ°μ§ μμ κ² κ°μ§λ§)
- λν, λ³Έ λ Όλ¬Έμμλ λμ μ₯μ λ¬ΌκΉμ§ ν΄κ²°ν μ μλμ§μ λν λ Όμκ° μλ€. μ²μ λΉννμλ μ λ°± μ μκ°μ΄ λΉμ·νκ² λμκ³ , λμ μ₯μ λ¬Όμ΄ μκ²Όμ λμ μκ°μ μΈ μ°μ°μλκ° ννΌνκΈ°μ μΆ©λΆνμ§λ κΆκΈνλ€.
μ 리 λμ.