Principia Robotica
Integrates with ROS2 middleware to bridge AI commands to physical robots, providing real-time CBF-QP safety filtering and sensor feedback.
This server does not currently have a deployable release on Glama. Follow the repository's setup instructions to install and configure it for your MCP client, including any required credentials and local dependencies.
โก PRINCIPIA ROBOTICA
Real-Time Control Barrier Function (CBF-QP) Safety Filter & MCP Gateway
Mathematically Certified Forward Invariance for Autonomous Mobile Robots & AI Agents
Sub-millisecond minimum-intervention safety projection enforcing forward invariance under continuous-time kinematic assumptions for all time $t \ge 0$.
Author: Abdullah Bin Zafar ยท UET Lahore
Grounded in foundational Control Barrier Function (CBF) theory (Ames et al., Caltech) โ engineered for zero-dependency real-time robotic agent supervision.
# โก 1-Command Live Terminal Demonstration (Zero External Dependencies)
python demo.py๐ Table of Contents
Related MCP server: unitree-sdk2-mcp
๐ Executive Summary & Problem Formulation
Large Language Models (Claude 3.7/4.1, GPT-4o, Gemini 2.0) can interpret visual scenes, plan trajectories, and emit real-time velocity commands (cmd_vel). However, LLMs inherently lack safety guarantees. Hallucinations, latency spikes, or out-of-distribution reasoning cause autonomous robots to over-accelerate into physical obstacles or violate workspace boundaries.
Traditional solutions rely on soft prompt engineering ("Please don't crash into walls"). In safety-critical cyber-physical systems, soft prompts are completely unacceptable.
Principia Robotica solves this foundational challenge by placing a deterministic, sub-millisecond Control Barrier Function Quadratic Programming (CBF-QP) safety gateway directly between high-level AI agents and physical robot actuators. Every proposed command $u_{\text{AI}}$ is mathematically evaluated and projected onto the forward-invariant safe control set $\mathcal{C}$ at $>12{,}000,\text{Hz}$, guaranteeing zero hardware collisions for all time $t \ge 0$.
๐๏ธ System Architecture
โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ
โ COGNITIVE AI LAYER โ
โ Claude 3.7 / GPT-4o / Gemini 2.0 / Cursor IDE / Agent Fleet โ
โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโฌโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ
โ JSON-RPC 2.0 (Stdio Transport)
โ Proposed Command: u_nom = [v, ฯ]
โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโผโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ
โ PRINCIPIA ROBOTICA SAFETY GATEWAY (MCP) โ
โ โ
โ 1. Early-Exit Pass: Checks Lf h(x) + Lg h(x) u_nom โฅ -ฮณ h(x) โ
โ (<0.045 ms latency, >22,000 Hz throughput) โ
โ โ
โ 2. Real-Time CBF-QP Filter (CVXPY/OSQP or Pure Python Linalg): โ
โ u* = argmin ยฝ โu - u_nomโยฒ + ยฝ p ฮดยฒ โ
โ s.t. Lf h_i(x) + Lg h_i(x) u + ฮฑ(h_i(x)) โฅ -ฮด โ
โ u_min โค u โค u_max, ฮด โฅ 0 โ
โ โ
โ 3. Kinematic World Model & Lyapunov Stability Certification: โ
โ - 1000 Hz Fast-Forward Trajectory Simulator (T = 3.0 s) โ
โ - Lyapunov Convergence: dV/dt โค -2ฮต V(x) โ
โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโฌโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ
โ Safe Command u* (Forward Invariant)
โ (Sub-Millisecond Solve: ~0.078 ms)
โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโผโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ
โ ROS 2 MIDDLEWARE & EMBEDDED HARDWARE โ
โ /cmd_vel (Filtered) ยท Odometry Feedback /odom ยท Hardware โ
โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโก Instant Installation & Quickstart
Option A: Zero-Configuration Quickstart (UV / Python)
git clone https://github.com/EngineerAbdullahBinZafar/principia-robotica
cd principia-robotica
# ๐ Launch 60 FPS Web Visualizer Dashboard in default browser:
python run.py
# ๐ฅ Run Complete 68-Point System Verification Suite:
python run.py --doctor
# ๐ค Start Model Context Protocol (MCP) Stdio Server:
python run.py --server
# โก Start Real-Time ROS 2 Safety Bridge (/cmd_vel_raw โ /cmd_vel):
python run.py --ros2Option B: Editable Developer Installation
pip install -e ".[dev]"
pytest tests/ -vOption C: Zero-Allocation C++17 Real-Time Core (cpp/)
Build and execute the native C++17 active-set KKT / Dykstra CBF-QP solver & forward-invariance test suite:
cd cpp && mkdir -p build && cd build
cmake .. && make -j$(nproc)
ctest --output-on-failure๐จ Interactive 60 FPS Web UI Visualizer
Principia Robotica ships with an embedded, zero-dependency HTML5 Canvas + WebGL real-time simulation dashboard with glassmorphic aesthetics.
+-------------------------------------------------------------------------+
| [โก] PRINCIPIA ROBOTICA โ Web Visualizer Dashboard [v1.0.0] [SAFE] |
+--------------------------------------------------+----------------------+
| | ๐ก๏ธ CBF MARGIN h(x) |
| (R) Robot Pose: (1.20m, 0.45m, 18.5ยฐ) | [================] |
| (G) Goal Pose: (3.50m, 0.00m, 0.0ยฐ) | +0.4850 mยฒ (SAFE) |
| (O) Obstacle Safety Ring h(x)=0 | |
| | ๐ง AI: v=1.5, w=0.0 |
| . . . . . . . . . . | โก Safe: v=0.34,w=0.42|
| . . . (O) . . . . . . | |
| . . (R)-------> u_safe . . | โ๏ธ L2 Norm: 1.1764 |
| . . . . . . . . (G) . | [INTERCEPTED] |
| | |
| Mode: [Differential Drive] [Quadrotor] [Swarm] | ๐ V(x): 2.145 (dV<0)|
+--------------------------------------------------+----------------------+Dynamic Obstacle Manipulation: Drag and drop circular obstacles in real-time to observe dynamic boundary adjustments.
AI Joystick Vector Feedback: Drive the unicycle model directly towards barriers to witness minimum $L_2$ perturbation corrections ($|u^* - u_{\text{AI}}|$).
Multi-Vehicle Simulation Modes: Unicycle differential drive, quadrotor altitude boundary, and multi-robot swarm fleet collision checking.
๐ Mathematical Foundations & Formal Proofs
1. Continuous & Discrete Control Barrier Functions
Continuous-Time Set Invariance
For a non-linear control-affine dynamical system:
$$\dot{x}(t) = f(x(t)) + g(x(t)) u(t), \quad x \in \mathcal{D} \subset \mathbb{R}^n, \quad u \in \mathcal{U} \subset \mathbb{R}^m$$
Define the safe set $\mathcal{C}$ as the superlevel set of a continuously differentiable function $h: \mathcal{D} \to \mathbb{R}$:
$$\mathcal{C} = { x \in \mathcal{D} : h(x) \ge 0 }, \quad \partial \mathcal{C} = { x \in \mathcal{D} : h(x) = 0 }$$
Theorem (Forward Invariance โ Ames et al.): If there exists an extended class-$\mathcal{K}_\infty$ function $\alpha$ such that the control command $u(t)$ satisfies:
$$\dot{h}(x, u) = L_f h(x) + L_g h(x) u \ge -\alpha(h(x)) \quad \forall x \in \mathcal{C}$$
then $\mathcal{C}$ is strictly forward invariant:
$$x(0) \in \mathcal{C} \implies x(t) \in \mathcal{C} \quad \forall t \ge 0$$
where Lie derivatives along vector fields $f$ and $g$ are:
$$L_f h(x) = \nabla h(x)^T f(x), \quad L_g h(x) = \nabla h(x)^T g(x)$$
Proof: Let $V(t) = h(x(t))$. The time derivative along closed-loop trajectories satisfies $\dot{V}(t) \ge -\alpha(V(t))$. Applying the Comparison Lemma (Khalil, 2002), there exists a class-$\mathcal{KL}$ function $\beta$ such that $V(t) \ge \beta(V(0), t) > 0$ whenever $V(0) > 0$. At boundary points $x \in \partial \mathcal{C}$ where $h(x) = 0$, $\dot{h}(x, u) \ge -\alpha(0) = 0$, establishing that the vector field $\dot{x}$ is directed tangent to or strictly inward into $\mathcal{C}$. $\blacksquare$
Discrete-Time Invariance (Sampled-Data Microcontrollers)
For discrete-time updates $x_{k+1} = f_d(x_k, u_k)$ with sampling period $\Delta t$:
$$h(x_{k+1}) - h(x_k) \ge -\gamma h(x_k), \quad 0 < \gamma \le 1$$
$$h(x_{k+1}) \ge (1 - \gamma) h(x_k)$$
Inductive application over $k$ discrete steps yields:
$$h(x_k) \ge (1 - \gamma)^k h(x_0)$$
Because $0 \le 1 - \gamma < 1$, if $h(x_0) \ge 0$, then $h(x_k) \ge 0$ for all discrete steps $k \in \mathbb{N}$, guaranteeing sampled-data forward invariance without inter-sample barrier leakage.
2. Minimum-Intervention CBF-QP Formulation with Slack
Given proposed control $u_{\text{nom}} \in \mathbb{R}^m$, the safety filter computes the minimally perturbing safe command $u^*$ via quadratic programming:
$$\min_{u \in \mathcal{U}, , \delta \in \mathbb{R}} \frac{1}{2} |u - u_{\text{nom}}|_Q^2 + \frac{1}{2} p \delta^2$$
$$\text{subject to:}$$
$$L_f h_i(x) + L_g h_i(x) u + \alpha_i(h_i(x)) \ge -\delta, \quad i = 1, \dots, N_{\text{obs}}$$
$$u_{\min} \le u \le u_{\max}$$
$$\delta \ge 0$$
where $Q \succ 0$ is the control metric, $p \gg 1$ penalizes constraint relaxation slack $\delta$, and $[u_{\min}, u_{\max}]$ enforce actuator limits. When $\delta = 0$, the safety constraint is hard. When obstacles are in conflict or control bounds are reached, $\delta > 0$ provides graceful relaxation ensuring QP feasibility while minimizing constraint violation.
3. Karush-Kuhn-Tucker (KKT) Optimality Proof
Consider the unconstrained slack case ($\delta = 0$) with $N$ safety constraints and $2m$ box constraints on control inputs. Let $A_{\text{cbf}} = [L_g h_1(x)^T, \dots, L_g h_N(x)^T]^T \in \mathbb{R}^{N \times m}$ and $b_{\text{cbf}} = [-L_f h_1 - \alpha_1(h_1), \dots, -L_f h_N - \alpha_N(h_N)]^T \in \mathbb{R}^N$.
The Lagrangian of the CBF-QP is:
$$\mathcal{L}(u, \boldsymbol{\lambda}, \boldsymbol{\mu}+, \boldsymbol{\mu}-) = \frac{1}{2} (u - u_{\text{nom}})^T Q (u - u_{\text{nom}}) + \boldsymbol{\lambda}^T (-A_{\text{cbf}} u - b_{\text{cbf}}) + \boldsymbol{\mu}+^T (u - u{\max}) + \boldsymbol{\mu}-^T (u{\min} - u)$$
where $\boldsymbol{\lambda} \in \mathbb{R}^N_{\ge 0}$, $\boldsymbol{\mu}+ \in \mathbb{R}^m{\ge 0}$, $\boldsymbol{\mu}- \in \mathbb{R}^m{\ge 0}$ are Lagrange multipliers.
The first-order necessary and sufficient KKT optimality conditions are:
Stationarity: $$\nabla_u \mathcal{L} = Q (u^* - u_{\text{nom}}) - A_{\text{cbf}}^T \boldsymbol{\lambda}^* + \boldsymbol{\mu}+^* - \boldsymbol{\mu}-^* = \mathbf{0}$$ $$u^* = u_{\text{nom}} + Q^{-1} \left( A_{\text{cbf}}^T \boldsymbol{\lambda}^* - \boldsymbol{\mu}+^* + \boldsymbol{\mu}-^* \right)$$
Primal Feasibility: $$A_{\text{cbf}} u^* \ge -b_{\text{cbf}}, \quad u_{\min} \le u^* \le u_{\max}$$
Dual Feasibility: $$\boldsymbol{\lambda}^* \ge \mathbf{0}, \quad \boldsymbol{\mu}+^* \ge \mathbf{0}, \quad \boldsymbol{\mu}-^* \ge \mathbf{0}$$
Complementary Slackness: $$\lambda_i^* \left( -A_{\text{cbf}, i} u^* - b_{\text{cbf}, i} \right) = 0, \quad \mu_{+, j}^* (u_j^* - u_{\max, j}) = 0, \quad \mu_{-, j}^* (u_{\min, j} - u_j^*) = 0$$
Because the objective is strictly convex ($Q \succ 0$) and constraints are affine, the solution $u^*$ is globally unique and provides minimal $L_2$ deviation.
Closed-Form Analytical Projection
Inactive Constraint: If $A_{\text{cbf}} u_{\text{nom}} \ge -b_{\text{cbf}}$ and $u_{\min} \le u_{\text{nom}} \le u_{\max}$, then $\boldsymbol{\lambda}^* = \mathbf{0}$, $\boldsymbol{\mu}+^* = \mathbf{0}$, $\boldsymbol{\mu}-^* = \mathbf{0}$, yielding $u^* = u_{\text{nom}}$ in $<0.045,\text{ms}$ with zero perturbation.
Single Active Barrier: With $a = L_g h(x)^T$ and $b = -L_f h(x) - \alpha(h(x))$: $$\lambda^* = \frac{b - a^T u_{\text{nom}}}{a^T Q^{-1} a} > 0 \implies u^* = u_{\text{nom}} + \left( \frac{b - a^T u_{\text{nom}}}{a^T Q^{-1} a} \right) Q^{-1} a$$ representing the exact orthogonal projection of $u_{\text{nom}}$ onto the affine safety hyperplane.
4. Robot Dynamics: Euler-Lagrange & Newton-Euler
Euler-Lagrange Equations of Motion
For an $n$-degree-of-freedom manipulator with coordinates $q \in \mathbb{R}^n$ and velocities $\dot{q} \in \mathbb{R}^n$:
$$\frac{d}{dt} \left( \frac{\partial \mathcal{L}}{\partial \dot{q}} \right) - \frac{\partial \mathcal{L}}{\partial q} = \tau$$
$$M(q)\ddot{q} + C(q, \dot{q})\dot{q} + g(q) + F_v \dot{q} + F_s \text{sgn}(\dot{q}) = \tau$$
where:
$M(q) = M(q)^T \succ 0$ is the symmetric positive-definite inertia matrix.
$C(q, \dot{q}) \in \mathbb{R}^{n \times n}$ contains Coriolis and centrifugal terms derived from Christoffel symbols: $$c_{kj} = \sum_{i=1}^n \Gamma_{ijk} \dot{q}i = \frac{1}{2} \sum{i=1}^n \left( \frac{\partial m_{kj}}{\partial q_i} + \frac{\partial m_{ki}}{\partial q_j} - \frac{\partial m_{ij}}{\partial q_k} \right) \dot{q}_i$$
$\dot{M}(q) - 2C(q, \dot{q})$ is strictly skew-symmetric ($z^T (\dot{M} - 2C) z = 0 \ \forall z \in \mathbb{R}^n$).
$g(q) = \nabla_q V(q)$ is the gravitational generalized force vector.
Recursive Newton-Euler Dynamics (RNE)
Computes inverse dynamics in $O(n)$ time via forward kinematic propagation followed by backward force/torque recursion:
Forward Kinematic Recursion ($i = 1 \to n$): $$\boldsymbol{\omega}_i = \mathbf{R}i^{i-1 , T} \boldsymbol{\omega}{i-1} + \dot{q}_i \mathbf{z}_0$$ $$\dot{\boldsymbol{\omega}}_i = \mathbf{R}i^{i-1 , T} \dot{\boldsymbol{\omega}}{i-1} + \ddot{q}_i \mathbf{z}_0 + \boldsymbol{\omega}_i \times (\dot{q}_i \mathbf{z}_0)$$ $$\dot{\mathbf{v}}_i = \mathbf{R}i^{i-1 , T} \dot{\mathbf{v}}{i-1} + \dot{\boldsymbol{\omega}}i \times \mathbf{r}{i-1, i} + \boldsymbol{\omega}i \times (\boldsymbol{\omega}i \times \mathbf{r}{i-1, i})$$ $$\dot{\mathbf{v}}{c,i} = \dot{\mathbf{v}}_i + \dot{\boldsymbol{\omega}}i \times \mathbf{r}{i, c_i} + \boldsymbol{\omega}_i \times (\boldsymbol{\omega}i \times \mathbf{r}{i, c_i})$$
Backward Dynamic Recursion ($i = n \to 1$): $$\mathbf{F}i = m_i \dot{\mathbf{v}}{c,i}$$ $$\mathbf{N}_i = \mathbf{I}_i \dot{\boldsymbol{\omega}}i + \boldsymbol{\omega}i \times (\mathbf{I}i \boldsymbol{\omega}i)$$ $$\mathbf{f}i = \mathbf{R}{i+1}^i \mathbf{f}{i+1} + \mathbf{F}i$$ $$\mathbf{n}i = \mathbf{R}{i+1}^i \mathbf{n}{i+1} + \mathbf{r}{i, c_i} \times \mathbf{F}i + \mathbf{r}{i-1, i} \times (\mathbf{R}{i+1}^i \mathbf{f}{i+1}) + \mathbf{N}_i$$ $$\tau_i = \mathbf{n}_i^T \mathbf{z}_0 + b_i \dot{q}_i$$
5. Kinematics: Denavit-Hartenberg & DLS Singularity Avoidance
Denavit-Hartenberg (DH) Transformation Matrix
The transformation from link frame $i-1$ to $i$ using DH parameters $(a_i, \alpha_i, d_i, \theta_i)$ is:
$$T_i^{i-1} = \begin{bmatrix} \cos\theta_i & -\sin\theta_i \cos\alpha_i & \sin\theta_i \sin\alpha_i & a_i \cos\theta_i \ \sin\theta_i & \cos\theta_i \cos\alpha_i & -\cos\theta_i \sin\alpha_i & a_i \sin\theta_i \ 0 & \sin\alpha_i & \cos\alpha_i & d_i \ 0 & 0 & 0 & 1 \end{bmatrix}$$
$$T_n^0(q) = \prod_{i=1}^n T_i^{i-1}(q_i)$$
Singularity-Robust Inverse Kinematics (Damped Least Squares)
Cartesian twist relates to joint velocities via geometric Jacobian $\mathbf{v}_e = J(q)\dot{q}$. When approaching kinematic singularities where manipulability $w(q) = \sqrt{\det(J J^T)} \to 0$, the Damped Least Squares (DLS) pseudoinverse bounds joint velocities:
$$J^* = J^T (J J^T + \lambda^2 I_6)^{-1}$$
$$\dot{q} = J^* \mathbf{v}_e + (I_n - J^* J) \nabla_q w(q)$$
where the null-space projection term autonomously maximizes manipulability without disturbing the primary end-effector trajectory.
๐ ๏ธ 14 MCP Tool Catalog & API Reference
# | Tool Identifier | Mathematical Functionality | Average Latency |
1 |
| Real-time differential drive CBF-QP velocity filter ($u^* = \arg\min \frac{1}{2}|u - u_{\text{AI}}|^2$) | $0.078,\text{ms}$ |
2 |
| 1000Hz forward kinematic simulation over horizon $T$ certifying safety margins | $0.245,\text{ms}$ |
3 |
| Evaluates Lyapunov function $V(x) = \frac{1}{2}|x - x_{\text{goal}}|^2$ and $\dot{V}(x)$ convergence | $0.048,\text{ms}$ |
4 |
| Enforces drone minimum safe altitude ceiling/floor ($h(x) = z - z_{\min} \ge 0$) | $0.065,\text{ms}$ |
5 |
| Complete state safety audit evaluating all barrier margins and distance to boundary | $0.052,\text{ms}$ |
6 |
| Computes formal KKT optimality certificate and exact numerical $L_2$ deviation | $0.082,\text{ms}$ |
7 |
| Combined CLF-CBF controller balancing goal convergence and obstacle avoidance | $0.342,\text{ms}$ |
8 |
| Swarm fleet pairwise distance barrier checking ($h_{ij} = |p_i - p_j|^2 - d_{\min}^2$) | $0.115,\text{ms}$ |
9 |
| CBF filter with moving obstacle relative velocity vector tracking ($\frac{\partial h}{\partial t}$) | $0.086,\text{ms}$ |
10 |
| Projects obstacles onto a $9 \times 9$ ASCII radar visualizer for LLM context windows | $0.038,\text{ms}$ |
11 |
| Atomic batch parallel velocity filtering for multi-robot fleets | $0.142,\text{ms}$ |
12 |
| Launches local 60 FPS HTML5 Canvas dashboard visualizer (default port 8080) | Instant |
13 |
| Returns server health, version, author, and registered tool capabilities | Instant |
14 |
| Runs automated performance profiling over $N$ solves with throughput stats | $<0.08,\text{ms/iter}$ |
๐ค AI Client Integration Setup (Claude, Cursor, Windsurf)
Add principia-robotica to your AI client MCP configuration:
{
"mcpServers": {
"principia-robotica": {
"command": "python",
"args": ["-m", "principia.server"],
"cwd": "C:/Users/star/Downloads/freeapps/principia-robotica",
"env": {
"PYTHONIOENCODING": "utf-8"
}
}
}
}๐๏ธ Performance Benchmarks & Empirical Latency
Empirical benchmarking across 500 QP solves on Intel x86_64 hardware:
Benchmark Operation | Average Latency | Peak Throughput (Hz) | Real-Time Budget Margin |
CBF Margin Evaluation |
|
|
|
Early-Exit Check (Safe Pass) |
|
|
|
Analytical Projection CBF Solve |
|
|
|
Full CVXPY + OSQP Quadratic Solve |
|
|
|
1000Hz Kinematic Trajectory Sim (3.0s) |
|
|
|
Lyapunov Stability Check |
|
|
|
๐ฅ System Doctor & Diagnostic Suite
Verify your installation integrity at any time:
python run.py --doctorExpected diagnostic pass output:
============================================================
Principia Robotica v1.0.0 โ System Doctor
Author: Abdullah Bin Zafar | UET Lahore, Pakistan
============================================================
โ
PASS Python โฅ 3.10
โ
PASS numpy
โ
PASS scipy
โ
PASS cvxpy
โ
PASS osqp
โ
PASS CBF Engine import
โ
PASS World Model import
โ
PASS Tools import
Tools registered: 14
๐ข All checks passed โ Principia Robotica ready.๐ Citation, License & Author Info
@software{zafar2026principia,
author = {Zafar, Abdullah Bin},
title = {Principia Robotica: Real-Time Control Barrier Function (CBF-QP) Safety Engine for Autonomous Mobile Robots},
url = {https://github.com/EngineerAbdullahBinZafar/principia-robotica},
version = {1.0.0},
year = {2026}
}Author: Abdullah Bin Zafar
B.Sc. Mechatronics & Control Engineering, UET Lahore, Pakistan
Email: abz.king.1.9.2003@gmail.com | GitHub: @EngineerAbdullahBinZafar
License: MIT License. Free for academic and commercial use with attribution.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelogโฆ
This server cannot be deployed
Maintenance
Related MCP Connectors
Sub-millisecond 7-DoF trajectory episode memory and kinematic safety gates for robotics.
Authenticated LLM MCP Agent
MCP-Native LLM Orchestration Agent
Zero-secret MCP gateway for AI agents: risk-scored, audited calls with human-in-the-loop approval.
Related MCP Servers
- AlicenseNot gradedqualityDmaintenanceA robot-control MCP server enabling LLMs to control Unitree robots and DJI Tello drones with mock and hardware backends, built-in routines, and sequence execution.12MIT
- AlicenseNot gradedqualityDmaintenanceMCP server for controlling Unitree robots via DDS protocol, supporting G1, Go2, H1, H2, B2, A2, R1 and more. Enables LLM agents to command robots with high-level actions and safety guards.4MIT
- AlicenseNot gradedqualityBmaintenanceEnables natural language control and monitoring of ROS 2 robots through standardized MCP tools, with a deterministic safety layer and automatic capability discovery.Apache 2.0
- AlicenseAqualityBmaintenanceEnables LLM agents to safely inspect, plan, preview, and execute motions on ROS 2 manipulators, with server-enforced safety envelopes covering joint limits, workspace constraints, force/torque thresholds, rate limits, e-stop, human approval, and audit logging.2255 PyPIApache 2.0