WATCH. QUESTION. CALCULATE.
A shuttle in the air. A moving chassis. A racket that has to arrive at the right place, at the right instant. Follow the ideas that connect them.
A project by Agasthya & Rajesh
Ten Robo-Badminton chapters, followed by three SOLIDWORKS practical lessons, in their original order.
Explore the physics ↓Learning material, with a live engineering checkpoint: these videos explain the project’s development. The current Q8 result passes the aggregate 90% finite single-return simulation test. Home recovery, continuous rallies and physical build qualification remain open; earlier videos may describe earlier layouts or targets.
01 / FOLLOW THE CHAPTERS
Loading chapters…
Select a chapter to begin. Your browser may require Play before automatic continuation.
02 / THE IDEAS TO TEST
Gravity pulls it down; air drag opposes its velocity. The current feather-flight model integrates these forces in small time steps instead of drawing a simple parabola.
dv/dt = −|v|v / L + g
Here v is velocity, L is the model’s drag length (4.49 m), and g points downward with magnitude 9.81 m/s². This is a simplified still-air model. Nylon needs its own measured model and validation.
Stereo vision compares where the shuttle appears in two calibrated cameras. In a simple parallel stereo pair, distance is approximately focal length × baseline ÷ disparity.
Z ≈ f B / d
The four moving heads use the usable pair at that instant. Their measured pan/tilt and the moving base pose matter: a smaller distance alone does not guarantee an accurate estimate. Occlusion, small disparity and motion blur can still cause losses.
New observations arrive while the robot is moving. Parallel preparation means the longer of the chassis and arm motions often controls readiness, rather than simply adding their durations.
Tavailable ≥ Tsense + Tplan + max(Tbase, Tarm)
This is a useful scheduling approximation; shared stability, collision constraints and camera acquisition can couple the motions. Q8 updates its supervisor every 10 ms. That is a command interval, not a claim that a wheel or arm reaches its new position in 10 ms.
Motors need enough force to accelerate the whole mass, and tyres need enough grip to transmit it. A basic level-ground traction estimate is a ≤ μg; stopping distance grows with speed squared.
F = ma · dstop ≈ v² / (2 abrake)
With the assumed μ = 0.7, the ideal translational bound is about 6.87 m/s². Actual wheel loads, steering, motor limits and the moving arm reduce what is available. Simulation checks these interactions; a real court grip test is still needed.
The racket must arrive with a useful face angle and relative velocity. The outgoing shuttle must clear the net and land in the opposite singles court. Some recorded contacts fail this second check.
vracket = vbase + ω × r + joint-motion contributions
The simulation uses a declared impact model and integrates outgoing flight. The replay includes misses and illegal contacts; touching the shuttle is not counted as a legal return. Stringbed behaviour and real contact parameters need experimental calibration.
A moving arm shifts loads on the wheels. A low centre of mass and suitable support width help resist tipping. After a return, chassis, arm and camera recovery must happen while the shuttle flies back.
Simple tipping estimate: a ≈ g b / h
Here b is support half-width in the acceleration direction and h is centre-of-mass height. It is a preliminary static estimate, not a certificate for this moving robot. Q8 currently records braking, then stops each independent attempt. Continuous home recovery/readiness has not yet passed a test.
03 / UNDERSTAND THE RESULT
1,838 / 2,000
Qualified legal returns in the frozen perturbed feeder test: 91.9%. The 420-attempt batch returned 391 shots (93.1%). These are independent starts using the same 21 feeder settings, not continuous rallies or all human shots.
Watch successes and failures on court ↗Keep the actual state after a hit. Recover toward a useful home region with the base and arm moving together, keep vision active, and respond to the human’s next return without resetting the robot.
Mechanical interfaces, masses, cables, actuator capability, structural checks and court safety also remain to qualify before purchase and construction.
The equations above describe the project’s declared computational models and basic mechanics. They are not new measurements of this robot.
The research team’s reported achievements belong to their robot and test protocol. This student design’s score comes from its own archived MuJoCo runs.