Competitive Task Allocation with Temporal Proposal Windows and Improved Artificial Potential Fields for Dual-Arm Industrial Manipulation

Surya Prakash S. K; Bhuvan Narula; Jagannath Prasad Sahoo; Jayakumar VK; Praful Hambarde; Amit Shukla

cs.ROOct 11, 2026

Competitive Task Allocation with Temporal Proposal Windows and Improved Artificial Potential Fields for Dual-Arm Industrial Manipulation

Authors: Surya Prakash S. K, Bhuvan Narula, Jagannath Prasad Sahoo, Jayakumar VK, Praful Hambarde, Amit Shukla

Organizations: School of Mechanical and Materials Engineering, IIT Mandi, Kamand, Mandi, 175005, H.P, India. · R&D, Addverb Technologies Limited, Sec-156, Phase-II, Noida, 201310, UP, India. · Centre for Artificial Intelligence and Robotics, IIT Mandi, Kamand, Mandi, 175005, H.P, India. · R&D 4IR, Dubai Electricity & Water Authority, , Dubai, P.O. Box 564, , UAE.

Abstract

Modern industrial automation demands autonomous collaborative systems capable of coordinated manipulation in workspaces with heterogeneous object arrangements. Existing systems rely on predetermined trajectories and rigid allocation strategies, causing suboptimal workspace utilization and collision risks. This paper presents a distance-based competitive task allocation framework for dual-arm systems, integrating real-time vision perception with collision-free motion planning. A temporal window mechanism enables each arm to evaluate object detections over five consecutive vision frames before committing to optimal distance-based selection, featuring dynamic reallocation and spatial conflict resolution at 0.01m threshold accuracy. An improved Artificial Potential Field (iAPF) framework provides coordinated dual-arm movements through exponential goal attraction, inter-arm repulsion, home-seeking, and velocity damping forces. Simulation results validate superiority over market-based, round-robin, and fixed allocation strategies in distance efficiency, collision risk, and workspace coverage. Hardware experiments using two 6-DOF Kinova Gen3 Lite manipulators demonstrate approximately 47% average velocity improvement over fixed allocation in homogeneous scenarios with zero collisions, while mixed object scenarios confirm universal handling with approximately 11% time overhead offset by superior safety margins. The framework integrates YOLOv8 OBB detection at 30 FPS with 10 Hz control loops, demonstrating practical deployment capability for industrial assembly and sorting operations.

Figures & tables

Figure 1: Overall Methodology: Intelligent Task Allocation for Dual Arm Efficient Motion Planning to Handle Objects.

Figure 2: Dual-Arm setup for intelligent task allocation

Predicted	Bolt	Nut	Reference	Background
Bolt	0.99	0.02	0.00	0.75
Nut	0.00	0.97	0.00	0.25
Reference	0.01	0.02	0.42	0.00
Background	0.00	0.00	0.57	0.00

Table 1: Normalized Confusion Matrix for YOLOv8 OBB Model (250 epochs)

Parameter	Value	Description
YOLOv8 inference rate	$>$ 70 FPS	RTX 3060 GPU
Camera acquisition rate	$\approx$ 30 FPS	RSD415 hardware limit at (640 $\times$ 480)
Control loop rate	10 Hz	Dual-arm coordination frequency
Detection confidence	0.75	YOLOv8 threshold (configurable)
ROS queue size	1	Latest-frame synchronization
Batch Size	1	Real-time single-frame processing

Table 2: Vision-control integration and task allocation parameters

Window Size	Total Distance (m)	Total Time (s)	Avg.Velocity (m/s)
1	$16.44\pm 0.32$	$145.24\pm 3.01$	$0.113\pm 0.0012$
3	$15.61\pm 0.16$	$132.87\pm 2.801$	$0.117\pm 0.0013$
5*	$\textbf{15.64}\pm\textbf{0.05}$	$\textbf{130.51}\pm\textbf{1.54}$	$\textbf{0.119}\pm\textbf{0.0012}$
7	$15.57\pm 0.13$	$144.72\pm 2.49$	$0.107\pm 0.0012$
10	$15.62\pm 0.07$	$160.22\pm 0.48$	$0.097\pm 0.0002$

Table 3: Temporal Window Ablation Results

Figure 3: Dual-arm workspace visualization comparing existing allocation methods with proposal method(our).

Figure 4: Three performance metrics quantify allocation quality. [(1) Distance Efficiency (1/m) computed as the inverse of mean object-to-arm distance, measuring workspace utilization where higher values indicate better proximity-based assignments; (2) Collision Risk, representing the number of detected trajectory intersections during simultaneous arm operation; and (3) Coverage (%), calculated as the percentage of successfully allocated objects relative to total objects.]

Figure 5: Force profile comparison showing iAPF advantages: Exponential attraction with adaptive gradient(left), Continuous repulsion eliminating traditional threshold discontinuity(right).

Figure 6: Dual-Arm Trajectories comparison with iAPF force(left) and with traditional APF(right).

Figure 7: Dual-Arm sequential movements with iAPF handling conflicting near-goal industrial components.

$C_{\text{goal}}$	Completion Dist. (m)	Completion Time (s)	Avg.Velocity (m/s)	Collisions	Overshoot
0.05	$15.50\pm 0.109$	$170.88\pm 0.32$	$0.090\pm 0.001$	N	N
0.10*	$\textbf{15.65}\pm\textbf{0.057}$	$\textbf{130.51}\pm\textbf{1.54}$	$\textbf{0.119}\pm\textbf{0.0012}$	N	N
0.15	$16.67\pm 0.138$	$119.64\pm 0.86$	$0.139\pm 0.0003$	N	Y
0.2	-	-	-	Y	-

Table 4:

C_{\text{goal}}

Parameter Validation (

C_{\text{obs}}

= 0.01) [Y: Yes N: No]

$C_{\text{obs}}$	Completion Dist. (m)	Completion Time (s)	Avg.Velocity (m/s)	Collisions	Overshoot
0.005	$13.87\pm 0.103$	$103.92\pm 3.89$	$0.133\pm 0.003$	N	N
0.010*	$\textbf{15.65}\pm\textbf{0.057}$	$\textbf{130.51}\pm\textbf{1.54}$	$\textbf{0.119}\pm\textbf{0.0012}$	N	N
0.015	$17.26\pm 0.172$	$143.38\pm 0.69$	$0.120\pm 0.0013$	N	Y
0.02	$18.46\pm 0.165$	$145.25\pm 0.57$	$0.127\pm 0.001$	N	Y

Table 5:

C_{\text{obs}}

Parameter Validation (

C_{\text{goal}}

= 0.1) [Y: Yes N: No]

Parameter	Value	Description
Attraction constant ( $k_{a}$ )	50,000	Exponential goal attraction gain
Repulsion constant ( $k_{r}$ )	3,000	Inter-arm repulsion gain
Home-seeking constant ( $k_{h}$ )	600,000	Home position attraction gain
Damping coefficient ( $k_{d}$ )	100	Velocity damping gain
Repulsion exponent ( $\eta$ )	2	Inverse-distance decay rate
Normalization factor ( $s$ )	4	Combined force scaling

Table 6: Implementation Parameters for the Improved Artificial Potential Field Framework

Figure 8: Dual-arm system sorting objects using dynamic task allocation. START: Both arms are at their home positions. (a) Both arms initiate simultaneous manipulation from home positions. (b-c) Coordinated motion planning—left arm (green trajectory) and right arm (red trajectory) execute independent pick-and-place operations. (d-e-f) Continued coordinated execution with iAPF collision avoidance maintaining safe inter-arm distances. END: Return to home positions. The sequence demonstrates zero collision events during compact object manipulation, validating the temporal proposal window mechanism and improved APF framework.

Figure 9: Dual-Arm system Analysis for Dynamic Task Allocation Single Object Type - Homogeneous

Figure 10: Dual-Arm system Analysis for Fixed Task Allocation Single Object Type - Homogeneous

Method	Scenario	Total Dist. (m)	Total Time (s)	Avg.Velocity (m/s)
Dynamic	Bolt-Only	$7.60\pm 0.082$	$65.134\pm 2.01$	$0.116\pm 0.003$
Dynamic	Nut-Only	$7.88\pm 0.1431$	$67.87\pm 1.06$	$0.116\pm 0.0002$
Fixed	Bolt-Only	$6.61\pm 0.016$	$84.65\pm 0.132$	$0.078\pm 0.0001$
Fixed	Nut-Only	$6.42\pm 0.0104$	$82.147\pm 0.08$	$0.0781\pm 0.0001$

Table 7: Performance Comparison: Dynamic vs Fixed Task Allocation.

Method	Scenario	Total Dist. (m)	Total Time (s)	Avg. Velocity (m/s)
Dynamic (Proposed)	Mixed	$15.49\pm 0.226$	$135.42\pm 0.485$	$0.114\pm 0.001$
Market-based	Mixed	$16.85\pm 0.315$	$152.10\pm 1.450$	$0.110\pm 0.002$
Round-robin	Mixed	$18.42\pm 0.412$	$175.30\pm 2.100$	$0.105\pm 0.003$
Fixed	Mixed	$12.57\pm 0.181$	$98.40\pm 0.800$	$0.128\pm 0.001$

Table 8: Mixed Object Scenarios: Hardware Comparison Across Allocation Methods (12 Objects, 10 Trials).

Figure 11: Combined Dual-Arm system Analysis Dynamic Task Allocation

Figure 12: Combined Dual-Arm system Analysis Fixed Task Allocation

Explore similar work

MAAP: Multi-Agent Active Perception for Collaborative Manipulation
Sep 18, 2026Bruno N. Y. Chen, Li Kang, Heng Zhou +4Active Vision for Robot Manipulation Robotic Manipulation
Carnegie Mellon University · Shanghai Jiao Tong University · University of Science and Technology of China +3
VAMP-MR: Vector-Accelerated Motion Planning and Execution for Multi-Robot-Arms
Jul 15, 2026Philip Huang, Chenrui Gao, Jiaoyang LiMulti-Robot Motion Planning Robotic Manipulation
Robotics Institute, Carnegie Mellon University, Pittsburgh, PA 15213, USA · University of Michigan, Ann Arbor
ToMPC: Task-oriented Model Predictive Control via ADMM for Safe Robotic Manipulation
Mar 14, 2026Xinyu Jia, Wenxin Wang, Jun Yang +2Robotic Manipulation Constrained Motion Planning
Department of Biomedical Engineering, National University of Singapore, Singapore 117583 · Peng Cheng Laboratory, Shenzhen 518057, China · School of Electrical and Electronic Engineering, Nanyang Technological University, Singapore 639798

Sep 18, 2026J/K move · Enter open · S save

MAAP: Multi-Agent Active Perception for Collaborative Manipulation

Bruno N. Y. Chen, Li Kang, Heng Zhou, Xiufeng Song+3

Carnegie Mellon University · Shanghai Jiao Tong University · University of Science and Technology of China+3

Active Vision for Robot Manipulation Robotic Manipulation Multi-Agent Systems Imitation Learning Robot Imitation Learning Multi-Robot Coordination

Open paper AskPDF arXiv

Multi-agent manipulation naturally produces multiple task-driven viewpoints: every arm carries a wrist camera and moves through the scene while acting. Yet these observations are typically underutilized, and active perception in manipulation is still often treated as requiring a dedicated sensing agent. We introduce MAAP (Multi-Agent Active Perception), in which every arm is dual-purpose: it executes manipulation actions and, through the wrist camera it carries, simultaneously serves as a moving viewpoint for the team. We pair this with RAIL (Role-Aware Imitation Learning), a controller that predicts each arm's current role alongside its action chunk and conditions action generation on it, representing role-dependent actions within one network. Across four simulated tasks, widening the perception regime lifts average success from 56.5% with a fixed camera to 62.5% with one active wrist view and 70.0% with all of them, while MAAP+RAIL reaches 79.2%. RAIL's additional gain is concentrated on the three-arm Microwave task, where success rises from 47% to 82% on identical multi-wrist inputs. On a dual-arm platform, MAAP+RAIL succeeds in 14 of 20 placement trials compared with 0 of 20 for fixed-view ACT. Collaborative manipulation can thus serve as an active perception mechanism in its own right.

Abstract

Explore similar work

MAAP: Multi-Agent Active Perception for Collaborative Manipulation

VAMP-MR: Vector-Accelerated Motion Planning and Execution for Multi-Robot-Arms

ToMPC: Task-oriented Model Predictive Control via ADMM for Safe Robotic Manipulation