M.Sc. student at Technion's TASP (Technion Autonomous Systems Program), focused on autonomous systems, multi-agent planning, and motion control. I bridge algorithmic rigor with practical implementation, combining theory, simulation, and real-world validation. Prior industry experience as Control & Automation Engineer at Unitronics, specializing in PLC/HMI programming, industrial protocols, and advanced motion systems.

Core Expertise: Multi-agent coordination • Motion planning • Reinforcement learning • Control systems • Industrial automation

• Environment-First Aggregate Transport — Novel planning paradigm for autonomous earthwork with trajectory texture fabric (Python, PyBullet, A*)
• Multi-Robot Waiter System — Task + motion planning with PDDL and RRT* for mobile manipulators
• UR5e 6-DOF Planning — RRT* in C-space, sim-to-real bead maze manipulation
• Motion Planning Algorithms — Minkowski sums, visibility graphs, Weighted A*, RRT/RRT* implementations
• Quadcopter RL Control — PPO/A2C flight stabilization in CoppeliaSim (350K+ timesteps)
• KDCR Control Systems — Inverse dynamics, robust control, adaptive control (5 strategies)
• Game AI & Machine Learning — Minimax, Alpha-Beta, ID3 decision trees, MDPs

Novel environment-centric planning paradigm for autonomous earthwork robotics

Publication Status:

• ✅ Accepted: ACM Symposium on Applied Computing (SAC) 2026 - IRMAS Track
• 📝 In preparation: Extended full paper for future submission

Designed and validated a groundbreaking approach to multi-agent coordination for construction robotics, shifting from traditional rover-centric to environment-centric perspective. The algorithm decouples environment representation from individual agent trajectories, enabling near-constant marginal cost for multi-robot scaling.

Autonomous earthwork concept: rovers push aggregates into target formations using environment-centric planning

Core algorithm: trajectory fabric with heat map (red=high flow corridors) guiding aggregate transport to target

2D→3D integration: PyBullet simulation with visibility cone constraints and curved pushing trajectory

Key Achievements:

• 🎯 Near-linear scaling: R² = 0.988 across 7,280 test iterations
• ⚡ 7-9x speedup through selective update mechanism
• 🔬 19.5% improvement from A* continuation reuse (Wilcoxon p=0.009)
• 📊 Memory efficient: ~410MB peak, 1.48MB average delta
• 🤖 Multi-agent ready: Framework designed for MAPF integration

Technical Contributions:

• Trajectory texture fabric with visibility cone constraints
• Heat map generation identifying high-flow corridors
• Spillage modeling via path curvature analysis
• Selective rebuild mechanism (affected region identification)
• 2D↔3D integration validation with PyBullet

Tech Stack: Python, NumPy, SciPy, Shapely, PyBullet, A* search, dynamic programming

📁 Repository | 📄 Full Paper (PDF)

RRT* motion planning for 4-DOF robotic arm manipulation

Autonomous multi-robot coordination combining task planning (PDDL) with motion planning (RRT/RRT)*

Built complete multi-robot system for autonomous restaurant service, coordinating two mobile manipulators to deliver dishes efficiently in dynamic environments.

Key Features:

• 🎯 Task Planning: PDDL-based task scheduling and order management
• 🗺️ Base Navigation: A* + PRM for collision-free 2D navigation
• 🦾 Arm Planning: RRT/RRT* for 4-DOF manipulation with iterative rewiring
• 🔄 Integration: Combined symbolic planning with geometric motion planning

Technical Implementation:

• Forward kinematics and collision detection for manipulator
• Nearest-neighbor search and tree rewiring logic for RRT*
• Multi-query planning for efficient path reuse
• End-effector path cost optimization

Tech Stack: Python, UPF (Unified Planning Framework), PDDL, RRT/RRT*, forward kinematics, matplotlib

📄 Repository

Real UR5e robot performing complex manipulation tasks with wooden blocks

Complete motion planning system for UR5e collaborative robot

Developed comprehensive robotic arm control system spanning kinematics, collision detection, and high-dimensional motion planning for complex manipulation tasks.

Key Achievements:

• 🎯 Forward/Inverse Kinematics: DH parameters with multiple solution handling
• 🔄 RRT Planning:* Asymptotically-optimal path planning in 6-DOF C-space
• 🧩 Complex Task: 3D bead-threading maze navigation with position + orientation constraints
• ⚙️ Collision Avoidance: Sphere-based multi-link collision detection

Project Highlights:

• Project 1-2: UR5e kinematics + RRT/RRT* global planning
• Project 3: Multi-waypoint reaching with geometric object arrangement (spelling "N" and "O")
• Project 4 (Capstone): 3D bead maze solving with Fréchet distance optimization

Technical Details:

• Numerical inverse kinematics solver with joint limit handling
• k-nearest neighbor rewiring for path cost minimization
• Task-space planning with orientation requirements
• Sim-to-real validation on actual UR5e hardware

Tech Stack: Python, NumPy, RRT/RRT*, DH parameters, 3D visualization, real UR5e validation

📄 Repository

Multi-target reaching (left) and inspection coverage planning (right)

Comprehensive implementation of exact and sampling-based planning algorithms

Implemented and compared fundamental motion planning algorithms from exact methods (visibility graphs) to sampling-based approaches (RRT*) with application-specific extensions.

Implemented Algorithms:

• 📐 Exact Planning: Minkowski sums, visibility graphs, Dijkstra shortest path
• 🎲 Sampling-Based: Weighted A* (grid), RRT, RRT* with asymptotic optimality
• 🎯 Application Planning: Multi-target reaching, inspection coverage optimization

Key Results:

• HW1: Minkowski sum C-space computation + visibility graph construction for diamond robot
• HW2: Comparative analysis of Weighted A* (ε tuning), RRT (goal bias sensitivity), RRT* convergence
• HW3: Task-specific planners with coverage metrics and path cost minimization

Performance Analysis:

• Success rate >95% with goal biasing
• RRT* converges to near-optimal solutions
• Parameter sensitivity analysis (goal probability 5%-20%, heuristic weights 1-20)

Tech Stack: Python, NumPy, Matplotlib, computational geometry, probabilistic roadmaps

📄 Repository

PPO agent learning autonomous flight control and stabilization

Model-free deep RL for autonomous quadcopter stabilization and navigation

Developed deep reinforcement learning agents (PPO, A2C) that learn to control a quadcopter in 3D space without explicit physics models or pre-programmed control laws.

System Architecture:

• 🎯 State Space (16D): Quaternion orientation, position, velocities, target
• 🔧 Action Space (4D): Propeller thrusts [0, 3.18825] N per motor
• 🎁 Reward Function: Multi-objective balancing stabilization, vertical tracking, horizontal distance

Key Challenges Solved:

• Python-CoppeliaSim Synchronization: Synchronous mode, blocking API calls, communication pause-resume
• Reward Shaping: Progressive curriculum learning across multiple training sequences
• Sample Efficiency: Algorithm comparison (PPO vs A2C), hyperparameter tuning

Training Results:

• ✅ Total Timesteps: 350,000+ across multiple sequences
• ✅ Network Architecture: [128, 128] neurons, discount γ=0.95
• ✅ Best Sequence: Sequence 3 (curriculum learning + PPO)

Tech Stack: Python, CoppeliaSim, Gymnasium, Stable-Baselines3, PPO/A2C, reward engineering

📄 Repository

Competitive agents using Minimax and Alpha-Beta Pruning in 5×5 warehouse game

Game-playing AI and supervised learning implementations

Implemented classical AI algorithms for adversarial search and machine learning, demonstrating mastery of fundamental decision-making techniques.

Game Search (Multi-Agent Environments):

• 🎯 Minimax: Full game-tree exploration with optimal move selection
• ⚡ Alpha-Beta Pruning: 8-10x speedup over Minimax while preserving optimality
• 🎲 Expectimax: Probabilistic opponent modeling for uncertain environments
• 🎨 Custom Heuristics: Multi-objective evaluation (score, resources, tactical positioning)

Machine Learning (Decision Trees & MDPs):

• 🌲 ID3 Algorithm: Information gain-based feature selection with entropy splitting
• 📊 Continuous Features: Dynamic threshold discretization for real-valued data
• 🔄 Cross-Validation: k-fold hyperparameter tuning (tree depth optimization)
• 🎯 MDPs: Value Iteration and Policy Iteration for sequential decision-making

Application: 5×5 warehouse game with competitive agents; medical diagnosis with 30 continuous features

Tech Stack: Python, NumPy, game tree search, decision trees, dynamic programming

📄 Repository

Comprehensive control systems: kinematics, dynamics, and 5 control strategies

Developed complete robotic control pipeline from geometric motion to dynamic behavior to adaptive control, with rigorous comparative analysis across multiple control methodologies.

System Components:

Kinematics:

• 🎯 Forward Kinematics: Homogeneous transformation matrices (4×4) for arbitrary serial/parallel manipulators
• 🔄 Inverse Kinematics: Analytical and numerical solutions with 8 solution branches (3-DOF arm)
• 📊 Jacobian Analysis: Full 6×6 matrices, singularity detection, force/torque mapping

Dynamics:

• ⚙️ Euler-Lagrange Formulation: Inertia matrices H(q), Coriolis/centrifugal C(q,q̇), gravity G(q)
• 📐 Parallel Robot Kinematics: 2D parallel manipulator with constraint equations

Control Strategies (Comparative Analysis):

1. Inverse Dynamics + PD: Model-based control with computed torque
2. PD + Gravity Compensation: Improved robustness without full dynamics
3. PID Control: Model-independent with integral action
4. MINMAX Robust Control: Handles bounded parametric uncertainty (load 0-0.5 kg)
5. Adaptive Control: Real-time parameter estimation via Lyapunov stability

Performance Results:

• ✅ Tracking Error: <0.15% nominal, <0.20% with 25% uncertainty
• ✅ Robustness: Adaptive maintained 99.5% performance under load variations
• ✅ Trade-offs: MINMAX highest robustness but 15-20% torque chattering; Adaptive best smoothness

Tech Stack: MATLAB (1500+ lines), Euler-Lagrange mechanics, Lyapunov stability, trajectory planning

📄 Repository

Multi-agent path planning • RRT/RRT* • A* • Dijkstra • Reinforcement learning (PPO, A2C) • MDP • Minimax • Alpha-Beta Pruning • Decision trees • Dynamic programming

Python (primary) • C/C++ (embedded) • MATLAB • ROS • PLC/HMI programming (ladder logic, structured text) • Git • NumPy • SciPy • Matplotlib

CoppeliaSim • PyBullet • Gazebo • Forward/inverse kinematics • Jacobian analysis • Collision detection • Trajectory planning

Motion control systems • Servo synchronization • EtherCAT real-time control • Modbus (RTU/TCP) • CANopen • BACnet/IP • MQTT • Ethernet/IP • SQL connectivity • Raw serial/TCP/CAN (Layer-2)

• Environment-First Planning for Aggregate Transport: Foundations for Multi-Agent Systems Nir Manor, Federico Oliva, Amir Degani ACM Symposium on Applied Computing (SAC) 2026 - IRMAS Track ✅ Accepted Extended full paper in preparation for future conference submission 📄 Full Paper (PDF)

Automation and control engineer with expertise in PLC/HMI programming, industrial communication protocols, and advanced motion systems. Design and implement full automation solutions from initial architecture through field deployment.

Key Contributions:

• ⚙️ Motion Control: Designed servo axis synchronization, EtherCAT-based real-time control, gear ratio mapping, encoder feedback scaling
• 🔌 Custom Interfaces: Built communication interfaces for unsupported devices using raw serial/TCP/CAN (Layer-2) messaging
• 🏗️ Complete Projects: Delivered PLC/HMI programming, device integration, system commissioning, legacy migrations
• 🔧 Troubleshooting: Diagnosed firmware regressions, communication instability, SQL faults, multi-protocol conflicts
• 📡 Protocol Expert: Modbus, CANopen, BACnet/IP, MQTT, Ethernet/IP, EtherCAT, SQL
• 👨‍🏫 Technical Training: Provided remote/on-site customer training on system design and troubleshooting
• 🔬 R&D Bridge: Prepared reproduction scenarios and structured Jira reports for issue resolution

• 📧 Email: il.nir.manor@gmail.com
• 💼 LinkedIn: linkedin.com/in/nir-manor-002225162
• 🎓 GitHub: github.com/NirManor