A multi-armed robot for assisting with agricultural tasks

Humans often use one hand to hold a branch for easier access, while the other hand is used to perform basic tasks such as (a) pruning branches and (b) hand pollinating the flower. (c) Overview of the approach used by Madhav and colleagues, where one robot manipulates the branch to move the flower into another robot’s field of view through force-aware path planning. Figure from Power of perceptive branch manipulation to aid farming tasks.

In their paper Power of perceptive branch manipulation to aid farming taskswhich was presented at IROS 2025, Madhav Men, Rashik Shrestha, Trevor Smithand Yoo Joe He proposed a methodology for safely handling branches to aid various agricultural tasks. We interviewed Madhav to find out more.

Can you give us an overview of the problem you address in the paper?

Madhav Men (MR): Our work is motivated by StickBug (1), a multi-arm robotic system for precision pollination in greenhouse environments. One of the main challenges that StickBug faces is that many flowers are partially or completely hidden within the plant canopy, making them difficult to detect and access directly for pollination. This challenge also arises in other agricultural tasks, such as fruit harvesting, where target fruits may be obscured by surrounding branches and foliage.

To address this problem, we studied how a robot arm could safely manipulate branches so that these covered flowers could be brought into the field of view or into a workspace accessible to another robot arm. This is a difficult handling problem because plant branches are deformable, brittle, and vary greatly from one branch to another. In addition, unlike pick-and-place tasks, where objects move freely in space, branches remain attached to the plant, imposing additional constraints on movement during manipulation. If the robot moves a branch without observing these constraints and safety limits, it may use excessive force and damage the branch.

So, the fundamental problem we address in this paper is: How can a robot safely manipulate branches to reveal hidden flowers while remaining aware of interaction forces and minimizing damage?

What was your approach to dealing with the problem?

Mr: Our approach (2) combines motion planning that takes into account branch constraints with real-time force feedback.

First, we create a feasible processing path using a scheme based on the RRT* (rapid random tree exploration) algorithm in the workspace. The plan respects the branch’s geometric constraints and mission requirements. We model branches as deformable linear objects and use a geometric heuristic to determine the safest configurations to handle.

Then, during execution, we monitor the interaction force using a force sensor mounted on the processor. If the measured force exceeds a pre-determined safe threshold, the system will not continue on the same path. Instead, it replans the movement online and searches for an alternative path or target configuration that can reduce branch fatigue while still getting the job done.

So, the basic idea is that the robot doesn’t just plan for accessibility. It also adapts its movement based on the physical response of the branch during manipulation.

Madhav with the multi-armed pollination robot, StickBug.

What are the main contributions of your work?

Mr: The main contributions of our work are:

1. An engineering indicative model for branch processing that does not require setting branch-specific parameters or performing a physical inspection.
2. A motion planning strategy for branch manipulation that respects both workspace and branch constraints, using geometric heuristics to guide RRT* and incorporating online replanning based on force feedback.
3. Experimental demonstration demonstrating that force feedback-based movement planning can protect branches from excessive force during manipulation.
4. Generalization across different branch types, since the method is primarily based on branch geometry and can be adapted online to compensate for model inaccuracy.

Can you talk about the experiments you did to test this approach?

Mr: We evaluated the proposed method through a set of branch manipulation experiments using five different starting positions, all targeting a common target region. Each configuration was tested 10 times, resulting in a total of 50 trials. The experiment is considered successful if the robot makes the grasping point within 5 cm of the target point. For all experiments, the planning time limit was set at 400 s, and the allowable interaction force range was 40 N to 40 N. Across the 50 trials, 39 trials were successful and 11 failed, which corresponds to a success rate of approximately 78%. The average number of replanning attempts across all scenarios was 20.

Regarding force reduction, the results showed clear progress in safety. Constraint-aware planning reduced the manipulation force from more than 100 N to less than 60 N. Accordingly, online force-aware remapping reduced the force from approximately 60 N to below the required 40 N threshold. This indicates that safety awareness through geometric heuristics, which designs branches as deformable linear objects, combined with force-aware online replanning, can effectively reduce interaction forces during manipulation.

Overall, experiments show that the proposed framework enables safer handling of branches while maintaining mission feasibility. By combining branch constraint-aware planning with real-time force feedback, the robot can adapt its movement to reduce excessive force and reduce the risk of branch damage. These results highlight the value of force-aware planning for practical robotic manipulation in agricultural settings.

Do you have plans to expand this business?

Mr: Yes, there are several directions to expand this work.

One current limitation is the need to define a safe force threshold in advance. In practice, different types of branches require different force limits for safe manipulation. A major direction for future work is to automatically learn or estimate safe force thresholds from branch geometry or visual cues.

Another extension is improved comprehension point selection. Instead of just replanning after grasping, the system can also think ahead about the most appropriate grasping point so that the processing force required is reduced from the beginning.

We are also interested in designing a compatible gripper with a built-in force sensor that is more suitable for handling sensitive branches. In the long term, we plan to incorporate this method into a multi-arm agricultural robot, where one arm manipulates the branch and the other pollinates, prunes or harvests.

Overall, this work is developing agricultural robots that can effectively manipulate branches to support tasks such as harvesting, pruning, and pollination. By detecting fruits, cutting points and flowers hidden within the canopy, this capability could help overcome major barriers to wider adoption of robot-assisted agricultural techniques.

References

(1) Smith, Trevor, Madhav Regal, Christopher Touch, R. Michael Potts, Jared Byrd, R. Tyler Cook, Andy Chu, Jason Gross, and Yu Ju. Stickbug Design: A six-arm micro-pollination robot. In 2024 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pages 69–75. Institute of Electrical and Electronics Engineers, 2024.
(2) Men, Madhav, Rashik Shrestha, Trevor Smith, and Yu Jo, Power of perceptive branch manipulation to aid farming tasks. In 2025 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pages 1217–1222. IEEE, 2025.

About Madhav

Madhav Rijal holds a Ph.D. candidate in mechanical engineering at West Virginia University and works in agricultural robotics. His research combines motion planning, optimization, agent collaboration, and distributed decision making to develop robotic systems for precision pollination and other plant interaction tasks. His current work focuses on branch manipulation and safe robot operation in agricultural environments.

Tags: Eros