A comprehensive new review published in Cyborg and Bionic Systems argues that researchers should focus more on single-legged robots (SLRs) – especially hopping designs – to solve the toughest remaining hurdles in building fast, agile, and energy-efficient legged robots that can rival animals in rough terrain.
Traditional wheeled and tracked robots struggle on stairs, rubble, or soft ground. Legged robots such as Boston Dynamics’ Spot or ETH Zurich’s ANYmal already outperform them in many real-world scenarios, but they remain heavy, power-hungry, and mechanically complex. According to lead author Jinyuan Liu from Zhejiang University, the key to the next breakthrough lies in deliberately simplifying the problem: study one leg first.
“Compared to complete multi-legged robots, single-legged robots have far fewer parts and typically use a clean, periodic hopping gait,” Liu explains. “One hopping cycle captures almost all the essential dynamics that later appear in walking, trotting, or galloping quadrupeds. This makes SLRs an ideal testbed for new structures, actuators, models, and controllers.”
The 50-page review, co-authored by researchers from Zhejiang University and the State Key Laboratory of Fluid Power and Mechatronic Systems, systematically classifies decades of SLR research along four dimensions:
1. Mechanical design
- Telescopic legs (prismatic joint, like a pogo stick) – simple, great for basic validation.
- Articulated legs (multi-joint, animal-like) – subdivided into:
– Rigid (RALR)
– Parallel elastic (PEALR)
– Series elastic (SEALR) – best at absorbing landing shocks and recycling energy
– Variable-stiffness (VSELR) – most adaptable but heaviest and hardest to control.
A detailed performance-vs-complexity table shows that series-elastic designs currently offer the best trade-off for high-speed dynamic locomotion.
2. Modeling approaches
- SLIP (Spring-Loaded Inverted Pendulum) family – the gold-standard template that captures centre-of-mass behaviour with minimal parameters.
- Articulated reduced-order models – more accurate for specific hardware, easier to use in advanced controllers.
3. Control strategies
- Model-based (MPC, whole-body control, virtual model control) – interpretable and safe but computationally heavy and sensitive to model errors.
- Model-free (central pattern generators, reinforcement learning) – excel in simulation and high-DoF systems but suffer from sim-to-real gap and lack of guarantees.
The authors highlight current solutions to the notorious sim-to-real problem: domain randomization, high-fidelity physics engines, and privileged-learning techniques that give the policy extra sensor information during training that is removed at deployment.
4. Future roadmap toward “true bionic motion”
The paper calls for tightly coupled advances in:
- bio-inspired morphology (muscle-like variable impedance, spines, tails)
- lightweight fabrication (topology optimization, multi-material 3D printing)
- new materials (high-energy-density elastomers, shape-memory alloys, soft actuators)
- hybrid systems (jump-gliding, grasping feet, reaction wheels)
- next-generation AI control (large-scale planning, world models, foundation policies).
“Only when morphology, actuation, and intelligence evolve together will we close the gap to animal-like performance in the wild,” Liu concludes.
The open-access article “Bridging the Gap to Bionic Motion: Challenges in Legged Robot Limb Unit Design, Modeling, and Control” appeared in Cyborg and Bionic Systems on 19 August 2025.
DOI: 10.34133/cbsystems.0365
Full text: https://spj.science.org/doi/10.34133/cbsystems.0365
With Boston Dynamics, Deep Robotics, Unitree, and several Chinese teams all racing toward ever faster and more robust quadrupeds, the authors believe intensified research on simplified single-legged platforms could dramatically accelerate the entire field in the coming years.
