Publiée 23 juillet 2026
PhD Position F/M Mechatronic whole-body co-design of a quadruped robot integrating local compliance for locomanipulation tasks.
Inria
Talence, Nouvelle-Aquitaine 33400, France
CDI
Contexte et atouts du poste
The PhD thesis will take place within the A uctus team at the Inria center of the University of Bordeaux (Talence) . Th is thesis falls within the scope of the team's scientific axis on the design and control of robotic systems. The project will benefit from the hardware and software resources of the lab experimental platform. The student will be supervised throughout the PhD by Margot Vulliez and David Daney, researchers specializing in modeling and analysis of mechanisms, mechatronic design, and robot architecture optimization.
T his PhD is part of a collaborative research project with the Gepetto team at LAAS-CNRS (Toulouse). The PhD candidate will work closely with a second PhD student, primarily based and supervised at Gepetto, and focusing on the development of RL motion generation and simulation-driven co-design tools. T he two PhD projects will benefit from regular input s from researchers across both partner teams. Several visits at the Gepetto team are planned for the PhD student to strengthen this collaboration through hands-on experience with the simulation tool or experimental studies.
Mission confiée
Quadruped robots have seen rapid growth over the past decade, after a major breakthrough established with MIT Cheetah [Seok13, Katz19]. T hey provide robust mobility and remarkable agility in unstructured environments. T hey are almost systematically based on a similar rigid architecture, equipped with quasi direct drive pancake motors all positioned in the torso t o minimize leg inertia [Wensing17]. Such r igid structures favor model accuracy and control tractability [Wensing23], but they also limit impact absorption and prevent the robot from exploiting local flexibility to improve mobility. Although quadrupeds are mature, their performance is plateauing far below that of biological counterparts.
M any alternative mechatronic solutions, such as elastic actuators ( SEA [Pratt95, Verstraten16], PEA [Mettin10, Ding24], VSA [Vanderborght13] ) , tensegrity mechanisms [Motro03, Vimbert26], flexible links [Xu22], soft and passive joints [Collins05, Badri- Spröwitz13 ], have already been proposed in the literature to enhance energy efficiency, improve shock absorption, provide intrinsic compliance, or reduce the robot mass, assembly constraints, and fabrication cost. E mbedding such propositions into an active trunk/ spine [Ijspeert07, Khoramshahi13] or locally exploiting compliance through flexible paws [Catalano21, Badri-Spröwitz22] or tails [Libby12, Liu25], has shown to increase locomotion speed and acceleration capability [Fisher17, Li23], enhance impact absorption [Sato15, Hua25], or stabilize gaits [Eckert15, Sabelhaus18]. While effective on targeted benchmarks, they remain isolated and fail to integrate into a comprehensive design. T hese innovations are typically investigated at the component level, without considering whole-body interactions, due to the lack of a co-design methodology that incorporates robust whole-body control policies .
Th is PhD project will leverag e recent co-design approach es and m odern control methods such as RL to go beyond component-level innovations. I t aims at developing a new whole-body quadruped architecture that embodies the use of alternative mechatronic concepts, to improve overall system manipulation and dynamic locomotion capabilities, impact tolerance, and energy efficiency . It will explicitly account for control constraints t hrough realistic simulation s of the robot behavior on targeted locomanipulation tasks.
Principales activités
To push the limits of manipulation and high-dynamic locomotion in quadruped robots requires moving beyond conventional rigid legged platforms. This shift calls for the development of new morphologies based on w hole-body , multi-functional designs that exploit alternative mechanisms and distributed compliance. This PhD thesis aims to propose, model, and compare different mechatronic propositions through both simulation-based evaluation and experimental calibration . The goal is to develop a novel integrated quadruped design , by adopting a whole-body co-design approach, jointly optimizing mechanical architecture, actuation systems, and control policy.
The thesis will be structured in several complementary axes of research:
- Simulation-informed development of alternative mechanisms for locomanipulation. The first axis consists of developing different mechatronic solutions that could improve dynamic locomanipulation capabilities, by drawing on insights from biological systems. We envision integrating localized flexibility and damping components (flexible links or soft joints), elastic transmissions (SEA, PEA), alternative kinematic structures (active or passive spines, tails, necks, parallel bar mechanisms), tensegrity mechanisms and tension elements. The localization and characteristics of candidate compliance/tension elements and additional degrees of freedom will be iteratively evaluated, refined, and compared through RL-based motion simulation, based on performance criteria (e.g. motion-force-stiffness range, energy consumption). Through a synthesis methodology, mechanisms will be adjusted to reproduc e optimal temporal profiles of force, motion, and stiffness, extracted from RL-generated trajectories.
- Experimental c haracterization of the alternative mechanisms and real-to-sim calibration. We will prototype and characterize the proposed mechanisms with a dual goal: (i) validating and refining their dynamic models through parameter identification on dedicated test benches, including external measurement systems (encoders, force sensors). The refined models will serve in simulation to reduce the real-to-sim gap; (ii) estimating each mechanism's performance and its sensitivity to variation in design or manufacturing parameters, such as used material, prototyping technology, or dimensional tolerances. This sensitivity analysis will inform parameters and levels of noise used in RL policy randomization t o improve the robustness and reliability of policy training. It will also provide the minimal accuracy required for part fabrication, thereby relaxing unnecessary and costly precision tolerances on the design.
- W hole-body design conceptualization, co-optimization, and prototyping. This final axis will consolidate all technological bricks together into a co-designed whole-body architecture. The new quadruped will move away from conventional designs by tightly integrating manipulation and locomotion capabilities through a global leg-spine-neck dynamic optimization. B uilding on the previous outcomes, w e will first define the quadruped's kinematic structure, actuation type, sensing modalities, and flexibility distribution, associated with a set of design parameters and a parameterized CAD model. We will then employ available co-design approaches to jointly optimize the architecture, actuation, and control policy through simulation. T he optimized full quadruped design will be prototyped, and its motion-force capabilities will be experimentally evaluated on a prescribed workspace. Exp e rimental locomanipulation demonstrations will validate the prototype performance on t wo applications: o utdoor high-speed galloping and flying-object catching.
Compétences
The candidate should have solid skills in mechanical design, prototyping, robot modeling, robotic control, and programming (C++ or Python).
Some additional experiences in optimization or in motor/sensor integration would be appreciated.
Avantages
Rémunération
2300€ / month before taxs
The PhD thesis will take place within the A uctus team at the Inria center of the University of Bordeaux (Talence) . Th is thesis falls within the scope of the team's scientific axis on the design and control of robotic systems. The project will benefit from the hardware and software resources of the lab experimental platform. The student will be supervised throughout the PhD by Margot Vulliez and David Daney, researchers specializing in modeling and analysis of mechanisms, mechatronic design, and robot architecture optimization.
T his PhD is part of a collaborative research project with the Gepetto team at LAAS-CNRS (Toulouse). The PhD candidate will work closely with a second PhD student, primarily based and supervised at Gepetto, and focusing on the development of RL motion generation and simulation-driven co-design tools. T he two PhD projects will benefit from regular input s from researchers across both partner teams. Several visits at the Gepetto team are planned for the PhD student to strengthen this collaboration through hands-on experience with the simulation tool or experimental studies.
Mission confiée
Quadruped robots have seen rapid growth over the past decade, after a major breakthrough established with MIT Cheetah [Seok13, Katz19]. T hey provide robust mobility and remarkable agility in unstructured environments. T hey are almost systematically based on a similar rigid architecture, equipped with quasi direct drive pancake motors all positioned in the torso t o minimize leg inertia [Wensing17]. Such r igid structures favor model accuracy and control tractability [Wensing23], but they also limit impact absorption and prevent the robot from exploiting local flexibility to improve mobility. Although quadrupeds are mature, their performance is plateauing far below that of biological counterparts.
M any alternative mechatronic solutions, such as elastic actuators ( SEA [Pratt95, Verstraten16], PEA [Mettin10, Ding24], VSA [Vanderborght13] ) , tensegrity mechanisms [Motro03, Vimbert26], flexible links [Xu22], soft and passive joints [Collins05, Badri- Spröwitz13 ], have already been proposed in the literature to enhance energy efficiency, improve shock absorption, provide intrinsic compliance, or reduce the robot mass, assembly constraints, and fabrication cost. E mbedding such propositions into an active trunk/ spine [Ijspeert07, Khoramshahi13] or locally exploiting compliance through flexible paws [Catalano21, Badri-Spröwitz22] or tails [Libby12, Liu25], has shown to increase locomotion speed and acceleration capability [Fisher17, Li23], enhance impact absorption [Sato15, Hua25], or stabilize gaits [Eckert15, Sabelhaus18]. While effective on targeted benchmarks, they remain isolated and fail to integrate into a comprehensive design. T hese innovations are typically investigated at the component level, without considering whole-body interactions, due to the lack of a co-design methodology that incorporates robust whole-body control policies .
Th is PhD project will leverag e recent co-design approach es and m odern control methods such as RL to go beyond component-level innovations. I t aims at developing a new whole-body quadruped architecture that embodies the use of alternative mechatronic concepts, to improve overall system manipulation and dynamic locomotion capabilities, impact tolerance, and energy efficiency . It will explicitly account for control constraints t hrough realistic simulation s of the robot behavior on targeted locomanipulation tasks.
Principales activités
To push the limits of manipulation and high-dynamic locomotion in quadruped robots requires moving beyond conventional rigid legged platforms. This shift calls for the development of new morphologies based on w hole-body , multi-functional designs that exploit alternative mechanisms and distributed compliance. This PhD thesis aims to propose, model, and compare different mechatronic propositions through both simulation-based evaluation and experimental calibration . The goal is to develop a novel integrated quadruped design , by adopting a whole-body co-design approach, jointly optimizing mechanical architecture, actuation systems, and control policy.
The thesis will be structured in several complementary axes of research:
- Simulation-informed development of alternative mechanisms for locomanipulation. The first axis consists of developing different mechatronic solutions that could improve dynamic locomanipulation capabilities, by drawing on insights from biological systems. We envision integrating localized flexibility and damping components (flexible links or soft joints), elastic transmissions (SEA, PEA), alternative kinematic structures (active or passive spines, tails, necks, parallel bar mechanisms), tensegrity mechanisms and tension elements. The localization and characteristics of candidate compliance/tension elements and additional degrees of freedom will be iteratively evaluated, refined, and compared through RL-based motion simulation, based on performance criteria (e.g. motion-force-stiffness range, energy consumption). Through a synthesis methodology, mechanisms will be adjusted to reproduc e optimal temporal profiles of force, motion, and stiffness, extracted from RL-generated trajectories.
- Experimental c haracterization of the alternative mechanisms and real-to-sim calibration. We will prototype and characterize the proposed mechanisms with a dual goal: (i) validating and refining their dynamic models through parameter identification on dedicated test benches, including external measurement systems (encoders, force sensors). The refined models will serve in simulation to reduce the real-to-sim gap; (ii) estimating each mechanism's performance and its sensitivity to variation in design or manufacturing parameters, such as used material, prototyping technology, or dimensional tolerances. This sensitivity analysis will inform parameters and levels of noise used in RL policy randomization t o improve the robustness and reliability of policy training. It will also provide the minimal accuracy required for part fabrication, thereby relaxing unnecessary and costly precision tolerances on the design.
- W hole-body design conceptualization, co-optimization, and prototyping. This final axis will consolidate all technological bricks together into a co-designed whole-body architecture. The new quadruped will move away from conventional designs by tightly integrating manipulation and locomotion capabilities through a global leg-spine-neck dynamic optimization. B uilding on the previous outcomes, w e will first define the quadruped's kinematic structure, actuation type, sensing modalities, and flexibility distribution, associated with a set of design parameters and a parameterized CAD model. We will then employ available co-design approaches to jointly optimize the architecture, actuation, and control policy through simulation. T he optimized full quadruped design will be prototyped, and its motion-force capabilities will be experimentally evaluated on a prescribed workspace. Exp e rimental locomanipulation demonstrations will validate the prototype performance on t wo applications: o utdoor high-speed galloping and flying-object catching.
Compétences
The candidate should have solid skills in mechanical design, prototyping, robot modeling, robotic control, and programming (C++ or Python).
Some additional experiences in optimization or in motor/sensor integration would be appreciated.
Avantages
- Subsidized meals
- Partial reimbursement of public transport costs
- Possibility of teleworking and flexible organization of working hours
- Professional equipment available (videoconferencing, loan of computer equipment, etc.)
- Social, cultural and sports events and activities
- Access to vocational training
- Social security coverage
Rémunération
2300€ / month before taxs
