Pith. sign in

REVIEW 3 major objections 4 minor 25 references

Decentralized Nonlinear Model Predictive Control-Based Flock Navigation with Real-Time Obstacle Avoidance in Unknown Obstructed Environments

T0 review · 3 major / 4 minor · reviewed 2026-08-15 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read A distributed nonlinear model predictive controller can steer a leader-follower robot flock through an unknown, obstacle-strewn environment using only locally sensed point clouds, and the optimizer runs in real time on a Raspberry Pi 4.

desk verdict Solid engineering extension with real HIL timing data, but the obstacle-avoidance safety claim has a blind spot: directional filtering can discard points the robot can reach within the horizon. read the letter →

arxiv 2505.09434 v2 pith:5YQ3LGPM submitted 2025-05-14 cs.RO cs.SYeess.SY

Nuthasith Gerdpratoom , Kaoru Yamamoto This is my paper · ORCID
classification cs.ROcs.SYeess.SY
keywords nonlinearmodelpredictivecontrolflockingobstacleavoidancepointclouddistributedhardware-in-the-loopmulti-robotsystems
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

Flocking robots usually need a clean analytic description of obstacles, but real robots only have point clouds from LiDAR or depth cameras. This paper claims that the gap can be closed: a distributed nonlinear model predictive controller can take each follower's local point cloud, filter and downsample it, and turn the surviving points into obstacle-avoidance constraints over the prediction horizon. The scheme is validated in a Gazebo simulation of a leader and two Husky followers in an obstructed playpen, and in hardware-in-the-loop runs where the followers solve their NMPC on Raspberry Pi 4 boards with average solver times of 17.19 ms and 24.6 ms. If the claim holds, it shows that NMPC flock navigation is practical on low-cost embedded hardware in unknown environments.

What carries the argument

The central object is the processed point-cloud index set $O^t_i$ in Eq. (17), which converts raw perception into a handful of constraints compatible with NMPC. The pipeline has three stages: directional filtering (Eq. 14) keeps points whose dot product with the intended-travel vector $p^{t|t}_{b,i}$ is nonnegative; downsampling picks the closest point in each angular sector (2D) or voxel (3D); neighbor exclusion removes points that correspond to teammates. These constraints, $h(\cdot)\le 0$, are imposed at every horizon step, so the solver plans a trajectory that avoids all retained points rather than reacting to the nearest one. The PANOC/OpEn solver is the numerical machinery that makes this constraint set solvable within a 95 ms cut-off.

What would settle it

In the Gazebo environment, put a moving obstacle (or an extra robot) that enters a follower's sensing range only after its point cloud has been processed, and have it cross the predicted path; a collision or a violation of the safety distance $r_s$ would show that the static-obstacle assumption breaks the guarantee. Alternatively, force a U-turn into the region that directional filtering discarded; a collision there would show that discarding points behind the reference plane can remove safety-critical obstacles.

Watch

Extended reading notes

Core claim

The paper's central claim is that local sensor data alone is enough to keep an NMPC flock safe and connected in an unknown obstructed environment. Each follower builds the processed point-cloud set $O^t_i$ by keeping only points ahead of a reference plane normal to its intended travel direction, downsampling to the nearest point per sector (2D LiDAR) or voxel (3D depth data), and deleting points that belong to neighboring agents. Every retained obstacle point $m\in O^t_i$ then enters the optimal control problem as an inequality constraint $h(p^{t+k+1|t}_i, p^t_{m|i}) \le 0$ for each prediction step $k$, i.e., the predicted trajectory must stay outside the safety circle of each point. The authors report that the resulting optimization, solved with the PANOC method via the OpEn code generator, runs at 17.19 ms and 24.6 ms average solve times on Raspberry Pi 4 in HIL, and that the fleet maintains connectivity better than a vector-field-histogram baseline while navigating barriers, hydrants, dumpsters, and cones in Gazebo.

Load-bearing premise

The load-bearing premise is that the obstacles detected in the point cloud at time $t$ stay fixed for the entire prediction horizon, because the same sensed points are used to constrain every future predicted position.

Editorial extensions

If this is right

  • The obstacle-avoidance layer no longer requires a precomputed analytic map; each agent's LiDAR or depth sensor supplies everything the NMPC needs.
  • Because the same pipeline handles 2D sectors and 3D voxels, the method transfers from 2D LiDAR on ground robots to depth-camera or 3D-LiDAR setups.
  • Average solve times of 17.19 ms and 24.6 ms on Raspberry Pi 4 show that the whole flocking-plus-avoidance computation fits inside a 100 ms control loop on low-cost embedded hardware.
  • Solving time grows with the number of detected agents and constraints, so the authors note that smaller sampling times or larger fleets can hit the 95 ms cut-off and return suboptimal solutions.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • A corollary the paper does not spell out is that the safety constraint is only as fresh as the last point cloud: an obstacle that enters after directional filtering is invisible to the current solve, so deployment in human-inhabited spaces would likely need a fast re-planning trigger based on newly appearing points.
  • The directional filtering step assumes the current heading is a good proxy for where the trajectory will go; in tight turns the plane could be widened or replaced by a cone to keep more points, a change that would trade solve speed for safety margin.
  • Grouping processed points into object-level constraints, which the authors mention as future work, would reduce constraint count from roughly one per LIDAR point to a handful per object; if the average solve time scales with constraint count, this could enable larger fleets on the same Raspberry Pi hardware.
  • The HIL experiment uses only one leader and two followers, so the reported solve times do not yet show how the scheme behaves with dense neighborhoods; before promising scalability, one would want a measurement of solve time versus the number of detected neighbor agents.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 4 minor

Summary. This paper extends the authors' prior distributed NMPC flock-navigation framework by adding a local obstacle-avoidance strategy based on raw LiDAR point clouds. The main technical additions are a directional filtering step, a down-sampling step, and an obstacle-avoidance inequality constraint integrated into the NMPC problem. The authors evaluate the approach in Gazebo with one leader and two followers, compare it against a VFH-based reactive controller, and report a hardware-in-the-loop experiment in which the follower NMPC solvers run on Raspberry Pi 4 boards. The paper claims safe navigation through an unknown obstructed environment with real-time performance.

Significance. If the claims hold, the paper is a useful engineering contribution: it demonstrates that a distributed NMPC flocking controller with point-cloud-based obstacle constraints can run on low-cost embedded hardware and that directional filtering plus down-sampling keeps the solver within a 100 ms sampling period. The HIL experiment with average solve times below 25 ms is a concrete, reproducible result. However, the safety claim is not supported by the presented analysis: the directional filter can discard obstacle points that the unicycle robot could reach within the prediction horizon, and the evaluation is a single run with one environment and no statistical basis. The paper also provides no formal guarantees of collision avoidance or connectivity, so the headline claims should be revised to match the evidence.

major comments (3)
  1. [§3.1.1, Eq. (14) and §3.2, Eq. (19)] The directional filter retains only points q satisfying <p_t|t_b,i, p_t_b,q|i> >= 0, i.e., points in the forward half-plane of the reference plane. For the unicycle model (23) with v in [-0.1, 1.0] m/s and omega in [-8, 8] rad/s, the reachable set over the horizon T=10, dt=0.1 includes points behind that plane: the robot can reverse its heading in about 0.4 s. An obstacle point behind the plane at time t can therefore lie on the optimal predicted trajectory while being absent from the constraint set O_t_i in (19). This undermines the claim that the NMPC enforces obstacle avoidance. Please either revise the filter to retain all points in the horizon-reachable set, or provide a closed-loop counterexample test with an obstacle placed just behind the filtering plane and update the safety claims accordingly.
  2. [§4.2–4.3, Figs. 7–13] The evaluation consists of a single run in a single environment with three robots. The conclusion that the fleet can navigate safely through an unknown obstructed environment is therefore supported only by one demonstration. Please add multiple trials with varied obstacle layouts and initial conditions, and report success rates, minimum distances to obstacles, and error bars or other statistics. The VFH comparison in Fig. 9 is also based on one trajectory, so the comparative claim needs repeated runs before it can be considered established.
  3. [§3.2, Eq. (19) and §4.3] The obstacle constraint uses the point cloud sensed at time t for all k = 0,...,T-1, which assumes obstacles are static over the prediction horizon. For moving obstacles, the constraint is invalid. This limitation should be stated explicitly. In addition, the HIL experiment accepts suboptimal solutions returned at the 95 ms cutoff; the assertion that these were 'sufficiently effective' should be quantified, for example by reporting the frequency of cutoff events, constraint violations at the returned solution, or a comparison with converged solutions on the same problem instances.
minor comments (4)
  1. [§3.1.2, Eq. (15)] The down-sampling notation is unclear: the segments [is:is+fs] with is in {0, fs, 2fs, ..., ns fs} may not partition the full index set, and the formula for ns should be checked. Please specify how the final partial segment is handled.
  2. [§4.2] The sentence 'In the demo, each follower recognized up to two neighbors if they are in the sensor's range, regardless of the point cloud processing, which only simplifies the obstacle avoidance constraint' is confusing. Please rephrase to clarify the role of neighbor exclusion and the effect of point cloud processing on the constraint set.
  3. [§4.3] Figure 13 reports average and peak solve times, but no solver convergence statistics are given. Reporting the number of times the cutoff was hit and the final cost or constraint residuals would strengthen the HIL feasibility claim.
  4. [§4.2, Eq. (24)] The reference-trajectory optimization in Eq. (24) is not described in enough detail for reproducibility. Please provide the waypoint set, the values of qp, qu, qT, and the exact trajectory used in the simulations.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the paper's obstacle-avoidance constraints and real-time results are direct design/measurement results, not derived from their own outputs.

full rationale

The paper extends the authors' prior NMPC flocking formulation by adding point-cloud-based obstacle constraints. The core new elements are definitions: directional filtering (Eq. 14), down-sampling (Eq. 15-16), neighbor exclusion (Eq. 17), and a hard obstacle-avoidance inequality (Eq. 19) imposed on processed sensor data. These are direct constraints on predicted states, not fitted parameters renamed as predictions, and no derived quantity is equivalent to an input by construction. The references to Nag et al. (2022) and Nag and Yamamoto (2024) supply the baseline flocking NMPC framework, but the current paper's contribution—local obstacle avoidance from point clouds and HIL timing—does not reduce to those citations; the obstacle constraints are stated explicitly and the timing results are measured empirically. The directional-filter blind spot noted by skeptical readers (Eq. 14 discards points behind a plane even though the unicycle model can later turn toward them) is a potential correctness gap in the safety claim, not a circularity: the filter is an input-processing choice, and the obstacle constraint does not assume the filtered set is complete. Therefore, no circular step can be exhibited with the paper's equations, and the appropriate finding is no significant circularity.

Assumptions & free parameters 10 free parameters · 6 assumptions · 0 invented entities

The central claim rests on a kinematic model, a static world, and heuristic point cloud filtering, plus many hand-tuned weights and thresholds. No new physical entities are introduced. The most consequential free parameters are the safety radius and downsampling factor, which directly trade safety against computation.

free parameters (10)
  • safety distance r_s = not specified in text
    Used in Eq. 18 as obstacle radius; controls the size of the exclusion zone and directly affects safety and feasibility of the constraint.
  • down-sampling factor f_s = 4
    Equation (15); determines the sector width for 2D downsampling, balancing the number of obstacle constraints against computational load.
  • separation distance d_i,sep = not specified in text
    In hard and soft separation constraints (Eq. 5-6), it sets the minimum inter-agent distance and is critical for flock connectivity.
  • discount factor gamma = 0.8
    In objective function (2), prioritizes near-future predictions; affects control smoothness and closed-loop behavior.
  • soft separation penalty rho_sep = 20
    Weights the soft separation penalty in Eq. 7; balances feasibility against separation enforcement beyond T_sep.
  • alignment weight beta_i = 0.5
    In Eq. 10, sets the alignment weight for neighbors behind the agent; influences flocking cohesion.
  • cohesion/alignment static coefficient q_i,st = 0.5
    In Eq. 12, sets the baseline trade-off between cohesion and alignment.
  • cohesion/alignment dynamic coefficient c_i = 10
    In Eq. 12, scales how quickly the weighting changes with distance to the weighted average position.
  • input bounds = v in [-0.1, 1.0] m/s, omega in [-8, 8] rad/s
    Set in Section 4.1; constrain the control inputs and thereby limit the reachable trajectories.
  • leader gains K_v and K_psi = not specified in text
    Used in the leader controller (Eq. 25); affect the leader's tracking behavior and thus the reference for followers.
assumptions (6)
  • domain assumption Unicycle kinematic model as prediction model (Eq. 22-23)
    The NMPC uses a simplified kinematic model; the paper acknowledges the mismatch with true dynamics and delegates to low-level PID controllers in simulation.
  • domain assumption Static obstacle assumption: obstacle positions p^t_m|i are fixed over the prediction horizon (Eq. 19)
    The obstacle constraints use point cloud positions sensed at time t for all future predicted states; dynamic obstacles would invalidate the safety guarantee.
  • ad hoc to paper Directional filtering can safely discard points behind the reference plane (Eq. 14)
    The filter removes all points with negative dot product with the intended direction, which may include obstacles that become relevant if the robot later turns.
  • domain assumption Neighbor exclusion based on predicted positions (Eq. 17)
    The point cloud excludes neighboring agents using predicted positions; misidentification could cause false obstacle constraints or missed separation.
  • domain assumption The modified flocking rules from prior work (Nag et al. 2022, Nag and Yamamoto 2024) are valid
    The paper adopts the hierarchy, subjective weights, and cohesion/alignment trade-off without re-deriving or testing them independently.
  • ad hoc to paper Suboptimal solutions returned at the 95 ms cut-off are sufficiently effective
    The paper relies on the fact that the solver often does not converge within the allotted time, yet the control actions still produce safe trajectories.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Decentralized Nonlinear Model Predictive Control-Based Flock Navigation with Real-Time Obstacle Avoidance in Unknown Obstructed Environments." pith.science (2026). https://pith.science/paper/5YQ3LGPM

@misc{pith2026250509434,
  author       = {Pith},
  title        = {Pith review of: Decentralized Nonlinear Model Predictive Control-Based Flock Navigation with Real-Time Obstacle Avoidance in Unknown Obstructed Environments},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/5YQ3LGPM}},
  note         = {Machine review of arXiv:2505.09434}
}
read the original abstract

This work extends our prior work on the distributed nonlinear model predictive control (NMPC) for navigating a robot fleet following a certain flocking behavior in unknown obstructed environments with a more realistic local obstacle avoidance strategy. More specifically, we integrate the local obstacle avoidance constraint using point clouds into the NMPC framework. Here, each agent relies on data from its local sensor to perceive and respond to nearby obstacles. A point cloud processing technique is presented for both two-dimensional and three-dimensional point clouds to minimize the computational burden during the optimization. The process consists of directional filtering and down-sampling that significantly reduce the number of data points. The algorithm's performance is validated through realistic 3D simulations in Gazebo, and its practical feasibility is further explored via hardware-in-the-loop (HIL) simulations on embedded platforms.

Figures

Figures reproduced from arXiv: 2505.09434 by the authors.

Figure 1
Figure 1. The illustration of the reference plane in the Directional filtering process [PITH_FULL_IMAGE:figures/full_fig_p008_1.png] view at source ↗
Figure 2
Figure 2. Agent 2’s raw point cloud visualization (left). Agent 2’s point cloud visualization after employing Directional filtering (right). A red circle represents the weighted average among three robots, where the position of the leader agent (red triangle) is given more weight. For three-dimensional point clouds, where the data points can be gathered from a depth camera or 3D LiDAR, the perceived data is unstructured, unli… view at source ↗
Figure 3
Figure 3. displays the simulated Husky UGVs and environment in our Gazebo 3D simulation. The robots are attached with a UST10 simulated 2D LiDAR at their body center. The LiDAR has a default range of 5 meters, 360 degrees angular range starting from the robot heading, and rotates counterclockwise, with 720 sampling points per round in the robot’s body frame polar coordinate. The output information from LiDAR contains a time s… view at source ↗
Figures from the paper (11 more)
Figure 4
Figure 4. Figure 4: The visualization of laser scan point cloud data from UST10 simulated 2D LiDAR and the robots’ frames in RViz. trajectory within the horizon period T based on the instantaneous inputs. The predicted trajectory will be shared with the followers within the detection rang…
Figure 5
Figure 5. Figure 5: The schematic diagram of an agent i’s distributed NMPC with point cloud processing. 4.2 Navigate in the Obstructed Environment We tested the proposed optimal control formulation and algorithm by letting the leader agent track a reference trajectory in an obstructed env…
Figure 6
Figure 6. Figure 6: The reference trajectory is shown in red, and the lines connecting waypoints are shown in broken blue. wi for i = 1, . . . 7 are waypoints [PITH_FULL_IMAGE:figures/full_fig_p015_6.png]
Figure 7
Figure 7. Figure 7: The trajectories of 3 agents in an obstructed environment, where the blue trajectory is the leader’s and the rest are followers. 15 [PITH_FULL_IMAGE:figures/full_fig_p015_7.png]
Figure 8
Figure 8. Figure 8: Time-series control inputs with constraint boundary, shown as broken red lines [PITH_FULL_IMAGE:figures/full_fig_p016_8.png]
Figure 9
Figure 9. Figure 9: The plot of the deviation from the centroid of the two approaches. 16 [PITH_FULL_IMAGE:figures/full_fig_p016_9.png]
Figure 10
Figure 10. Figure 10: Schematic diagram of the HIL experimentation: The algorithm for the two followers is executed in each Raspberry Pi, while the station computer is responsible for the entire simulation. The information is shared among them using TCP/IP-based client-server communication…
Figure 11
Figure 11. Figure 11: HIL simulated trajectories of 3 agents in an obstructed environment, where the blue trajectory is the leader’s and the rest are followers [PITH_FULL_IMAGE:figures/full_fig_p018_11.png]
Figure 12
Figure 12. Figure 12: HIL simulation’s time-series of the control inputs with boundary, depicted in broken red. optimizer was analyzed, revealing peak values reaching the maximum solving time for a few samples, with an average of less than 25 ms. We acknowledge that if the sampling time ge…
Figure 13
Figure 13. Figure 13: NMPC’s solver solving time on Raspberry Pi 4B: agent 1 averages 17.19 ms and peaks at 95 ms, while agent 2 averages 24.6 ms and also peaks at 95 ms [PITH_FULL_IMAGE:figures/full_fig_p019_13.png]
Figure 14
Figure 14. Figure 14: The Raspberry Pi’s NMPC solver node’s CPU usage percentages for agents 1 and 2 are as follows: peak percentages of 116.2% and 117.3%, and average percentages of 98.2% and 97.82%, respectively. can group nearby points into a single object. This would reduce the number …

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

25 extracted references · 19 canonical work pages

  1. [1]

    Andersson, J. A. E., Gillis, J., Horn, G., Rawlings, J. B., and Diehl, M. (2019). CasADi: a software framework for nonlinear optimization and optimal control. Mathematical Programming Computation 11, 1–36. doi:10.1007/s12532-018-0139-4

    doi:10.1007/s12532-018-0139-4 2019

  2. [2]

    Barraquand, J., Langlois, B., and Latombe, J.-C. (1991). Numerical potential field techniques for robot path planning. In Fifth International Conference on Advanced Robotics ’Robots in Unstructured Environments. 1012–1017 vol.2

    work page 1991

  3. [3]

    and Stoynova, A

    Brayanov, N. and Stoynova, A. (2019). Review of hardware-in-the-loop – a hundred years progress in the pseudo-real testing. Electrotechnica & Electronica 54, 70–84

    work page 2019

  4. [4]

    Cao, H., Chen, J., Mao, Y ., Fang, H., and Liu, H. (2010). Formation control based on flocking algorithm in multi-agent system. In 2010 8th World Congress on Intelligent Control and Automation . 2289–2294 [Dataset] Clearpath Robotics (2015). Husky UGV tutorials 1.0.0. Available: https://clearpathrobotics.com

    work page 2010

  5. [5]

    Goarin, M., Li, G., Saviolo, A., and Loianno, G. (2024). Decentralized nonlinear model predictive control for safe collision avoidance in quadrotor teams with limited detection range. arXiv preprint arXiv:2409.17379 IEEE International Conference on Robotics and Automation (ICRA) 2025

    arXiv 2024

  6. [6]

    and Howard, A

    Koenig, N. and Howard, A. (2004). Design and use paradigms for Gazebo, an open-source multi-robot simulator. In 2004 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) (IEEE Cat. No.04CH37566). 2149–2154 vol.3

    work page 2004

  7. [7]

    Kong, F., Chen, H., Li, H., Yan, J., Wang, X., and Fang, J. (2023). Flocking with obstacle avoidance for fixed-wing unmanned aerial vehicles via nonlinear model predictive control. In 2023 42nd Chinese Control Conference (CCC). 5957–5962. doi:10.23919/CCC58697.2023.10240689

    arXiv 2023

  8. [8]

    Li, C., Yang, Y ., Jiang, G., and Chen, X. (2024). A flocking control algorithm of multi-agent systems based on cohesion of the potential function. Complex & Intelligent Systems 10, 2585–2604 20 Gerdpratoom et al. NMPC-Based Flocking with Local Obstacle Avoidance

    work page 2024

Show all 25 references
  1. [9]

    Liang, Q., Wang, Z., Yin, Y ., Xiong, W., Zhang, J., and Yang, Z. (2023). Autonomous aerial obstacle avoidance using lidar sensor fusion. Plos one 18, e0287177

    2023

  2. [10]

    S., Kanellakis, C., and Nikolakopoulos, G

    Lindqvist, B., Mansouri, S. S., Kanellakis, C., and Nikolakopoulos, G. (2020). Collision free path planning based on local 2d point-clouds for mav navigation. In 2020 28th Mediterranean Conference on Control and Automation (MED). 538–543

    2020

  3. [11]

    Lindqvist, B., Sopasakis, P., and Nikolakopoulos, G. (2021). A scalable distributed collision avoidance scheme for multi-agent UA V systems. In2021 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). 9212–9218

    2021

  4. [12]

    Mestres, P., Nieto-Granda, C., and Cort´es, J. (2024). Distributed safe navigation of multi-agent systems using control barrier function-based controllers. IEEE Robotics and Automation Letters 9, 6760–6767. doi:10.1109/LRA.2024.3414268 Mihaliˇc, F., Truntiˇc, M., and Hren, A. ...

    2024

  5. [13]

    Nag, A., Huang, S., Themelis, A., and Yamamoto, K. (2022). Flock navigation with dynamic hierarchy and subjective weights using nonlinear MPC. In 2022 IEEE Conference on Control Technology and Applications (CCTA). 1135–1140

    2022

  6. [14]

    and Yamamoto, K

    Nag, A. and Yamamoto, K. (2024). Distributed control for flock navigation using nonlinear model predictive control. Advanced Robotics 38, 619–631

    2024

  7. [15]

    Olfati-Saber, R. (2006). Flocking for multi-agent dynamic systems: algorithms and theory. IEEE Transactions on Automatic Control 51, 401–420. doi:10.1109/TAC.2005.864190

    2006

  8. [16]

    Reynolds, C. W. (1987). Flocks, herds and schools: A distributed behavioral model. SIGGRAPH Comput. Graph. 21, 25–34. doi:10.1145/37402.37406

    1987

  9. [17]

    Sathya, A., Sopasakis, P., Van Parys, R., Themelis, A., Pipeleers, G., and Patrinos, P. (2018). Embedded nonlinear model predictive control for obstacle avoidance using PANOC. In 2018 European Control Conference (ECC). 1523–1528. doi:10.23919/ECC.2018.8550253

    2018

  10. [18]

    and Luo, J

    Shi, L. and Luo, J. (2024). A framework of point cloud simplification based on voxel grid and its applications. IEEE Sensors Journal 24, 6349–6357

    2024

  11. [19]

    and Kumar, V

    Song, P. and Kumar, V . (2002). A potential field based approach to multi-robot manipulation. In Proceedings 2002 IEEE International Conference on Robotics and Automation (Cat. No.02CH37292) . 1217–1222 vol.2

    2002

  12. [20]

    Sopasakis, P., Fresk, E., and Patrinos, P. (2020). OpEn: Code generation for embedded nonconvex optimization. IF AC-PapersOnLine53, 6548–6554

    2020

  13. [21]

    Stella, L., Themelis, A., Sopasakis, P., and Patrinos, P. (2017). A simple and efficient algorithm for nonlinear model predictive control. In IEEE Conference on Decision and Control (CDC) . 1939–1944

    2017

  14. [23]

    Tanner, H., Jadbabaie, A., and Pappas, G. (2003b). Stable flocking of mobile agents part II: dynamic topology. In 42nd IEEE International Conference on Decision and Control (IEEE Cat. No.03CH37475) . vol. 2, 2016–2021. doi:10.1109/CDC.2003.1272911 21 Gerdpratoom et al. NMPC-Ba...

    2003 arXiv

  15. [24]

    Xu, T., Liu, J., Zhang, Z., Chen, G., Cui, D., and Li, H. (2023). Distributed mpc for trajectory tracking and formation control of multi-uavs with leader-follower structure. IEEE Access 11, 128762–128773. doi:10.1109/ACCESS.2023.3329232

    2023

  16. [25]

    Xu, Y ., Tong, X., and Stilla, U. (2021). V oxel-based representation of 3d point clouds: Methods, applications, and its potential use in the construction industry. Automation in Construction 126

    2021

  17. [26]

    Yu, S., Hirche, M., Huang, Y ., Chen, H., and Allg¨ower, F. (2021). Model predictive control for autonomous ground vehicles: a review. Autonomous Intelligent Systems 1, 1–17 FIGURE CAPTIONS Figure 1 The illustration of the reference plane in the Directional filtering process. ...

    2021

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.